diff --git a/.gitattributes b/.gitattributes new file mode 100644 index 0000000..ad20874 --- /dev/null +++ b/.gitattributes @@ -0,0 +1 @@ +model_data/yolov5_s_v6.1.pth filter=lfs diff=lfs merge=lfs -text diff --git a/.gitignore b/.gitignore index aa50ea1..5190304 100644 --- a/.gitignore +++ b/.gitignore @@ -1 +1,10 @@ -database/chestXray8_512/* \ No newline at end of file +database/chestXray8_512/* +database/* +logs/* + +__pycache__ +*/__pycache__ + +.vscode/ + +!*.md \ No newline at end of file diff --git a/README.md b/README.md index dba7f9a..29e25e7 100644 --- a/README.md +++ b/README.md @@ -5,5 +5,4 @@ # 数据集介绍 - -数据集来自kaggle[数据集原地址] +[详细见](./database/Dataset.md) diff --git a/config.ini b/config.ini index 8e7aa8a..89e240b 100644 --- a/config.ini +++ b/config.ini @@ -43,3 +43,6 @@ mixup_prob = 0.5 special_aug_ratio = 0.7 ; label_smoothing 标签平滑。一般0.01以下。如0.01、0.005。 label_smoothing = 0.01 + +[dataset] +classes_path = model_data/voc_classes.txt diff --git a/database/Dataset.md b/database/Dataset.md new file mode 100644 index 0000000..3afa003 --- /dev/null +++ b/database/Dataset.md @@ -0,0 +1,13 @@ +# 数据集介绍 + +数据集来自kaggle[数据集原地址](https://www.kaggle.com/datasets/spritan1/yolo-annotated-chestxray-8-object-detection) + +## 标签 +1. 肺不张，Atelectasis +2. 心脏肥大，Cardiomegaly +3. 很有效率，Effusion +4. 渗透，Infiltrate +5. 诺依，Nodule +6. 质量，Mass +7. 肺炎，Pneumonia +8. 气胸，Pneumothorax \ No newline at end of file diff --git a/database/README.md b/database/README.md index e69de29..175f18e 100644 --- a/database/README.md +++ b/database/README.md @@ -0,0 +1 @@ +# 存放数据集 \ No newline at end of file diff --git a/get_map.py b/get_map.py new file mode 100644 index 0000000..ccda5eb --- /dev/null +++ b/get_map.py @@ -0,0 +1,138 @@ +import os +import xml.etree.ElementTree as ET + +from PIL import Image +from tqdm import tqdm + +from utils.utils import get_classes +from utils.utils_map import get_coco_map, get_map +from yolo import YOLO + +if __name__ == "__main__": + ''' + Recall和Precision不像AP是一个面积的概念，因此在门限值（Confidence）不同时，网络的Recall和Precision值是不同的。 + 默认情况下，本代码计算的Recall和Precision代表的是当门限值（Confidence）为0.5时，所对应的Recall和Precision值。 + + 受到mAP计算原理的限制，网络在计算mAP时需要获得近乎所有的预测框，这样才可以计算不同门限条件下的Recall和Precision值 + 因此，本代码获得的map_out/detection-results/里面的txt的框的数量一般会比直接predict多一些，目的是列出所有可能的预测框， + ''' + #------------------------------------------------------------------------------------------------------------------# + # map_mode用于指定该文件运行时计算的内容 + # map_mode为0代表整个map计算流程，包括获得预测结果、获得真实框、计算VOC_map。 + # map_mode为1代表仅仅获得预测结果。 + # map_mode为2代表仅仅获得真实框。 + # map_mode为3代表仅仅计算VOC_map。 + # map_mode为4代表利用COCO工具箱计算当前数据集的0.50:0.95map。需要获得预测结果、获得真实框后并安装pycocotools才行 + #-------------------------------------------------------------------------------------------------------------------# + map_mode = 0 + #--------------------------------------------------------------------------------------# + # 此处的classes_path用于指定需要测量VOC_map的类别 + # 一般情况下与训练和预测所用的classes_path一致即可 + #--------------------------------------------------------------------------------------# + classes_path = 'model_data/voc_classes.txt' + #--------------------------------------------------------------------------------------# + # MINOVERLAP用于指定想要获得的mAP0.x，mAP0.x的意义是什么请同学们百度一下。 + # 比如计算mAP0.75，可以设定MINOVERLAP = 0.75。 + # + # 当某一预测框与真实框重合度大于MINOVERLAP时，该预测框被认为是正样本，否则为负样本。 + # 因此MINOVERLAP的值越大，预测框要预测的越准确才能被认为是正样本，此时算出来的mAP值越低， + #--------------------------------------------------------------------------------------# + MINOVERLAP = 0.5 + #--------------------------------------------------------------------------------------# + # 受到mAP计算原理的限制，网络在计算mAP时需要获得近乎所有的预测框，这样才可以计算mAP + # 因此，confidence的值应当设置的尽量小进而获得全部可能的预测框。 + # + # 该值一般不调整。因为计算mAP需要获得近乎所有的预测框，此处的confidence不能随便更改。 + # 想要获得不同门限值下的Recall和Precision值，请修改下方的score_threhold。 + #--------------------------------------------------------------------------------------# + confidence = 0.001 + #--------------------------------------------------------------------------------------# + # 预测时使用到的非极大抑制值的大小，越大表示非极大抑制越不严格。 + # + # 该值一般不调整。 + #--------------------------------------------------------------------------------------# + nms_iou = 0.5 + #---------------------------------------------------------------------------------------------------------------# + # Recall和Precision不像AP是一个面积的概念，因此在门限值不同时，网络的Recall和Precision值是不同的。 + # + # 默认情况下，本代码计算的Recall和Precision代表的是当门限值为0.5（此处定义为score_threhold）时所对应的Recall和Precision值。 + # 因为计算mAP需要获得近乎所有的预测框，上面定义的confidence不能随便更改。 + # 这里专门定义一个score_threhold用于代表门限值，进而在计算mAP时找到门限值对应的Recall和Precision值。 + #---------------------------------------------------------------------------------------------------------------# + score_threhold = 0.5 + #-------------------------------------------------------# + # map_vis用于指定是否开启VOC_map计算的可视化 + #-------------------------------------------------------# + map_vis = False + #-------------------------------------------------------# + # 指向VOC数据集所在的文件夹 + # 默认指向根目录下的VOC数据集 + #-------------------------------------------------------# + VOCdevkit_path = 'VOCdevkit' + #-------------------------------------------------------# + # 结果输出的文件夹，默认为map_out + #-------------------------------------------------------# + map_out_path = 'map_out' + + image_ids = open(os.path.join(VOCdevkit_path, "VOC2007/ImageSets/Main/test.txt")).read().strip().split() + + if not os.path.exists(map_out_path): + os.makedirs(map_out_path) + if not os.path.exists(os.path.join(map_out_path, 'ground-truth')): + os.makedirs(os.path.join(map_out_path, 'ground-truth')) + if not os.path.exists(os.path.join(map_out_path, 'detection-results')): + os.makedirs(os.path.join(map_out_path, 'detection-results')) + if not os.path.exists(os.path.join(map_out_path, 'images-optional')): + os.makedirs(os.path.join(map_out_path, 'images-optional')) + + class_names, _ = get_classes(classes_path) + + if map_mode == 0 or map_mode == 1: + print("Load model.") + yolo = YOLO(confidence = confidence, nms_iou = nms_iou) + print("Load model done.") + + print("Get predict result.") + for image_id in tqdm(image_ids): + image_path = os.path.join(VOCdevkit_path, "VOC2007/JPEGImages/"+image_id+".jpg") + image = Image.open(image_path) + if map_vis: + image.save(os.path.join(map_out_path, "images-optional/" + image_id + ".jpg")) + yolo.get_map_txt(image_id, image, class_names, map_out_path) + print("Get predict result done.") + + if map_mode == 0 or map_mode == 2: + print("Get ground truth result.") + for image_id in tqdm(image_ids): + with open(os.path.join(map_out_path, "ground-truth/"+image_id+".txt"), "w") as new_f: + root = ET.parse(os.path.join(VOCdevkit_path, "VOC2007/Annotations/"+image_id+".xml")).getroot() + for obj in root.findall('object'): + difficult_flag = False + if obj.find('difficult')!=None: + difficult = obj.find('difficult').text + if int(difficult)==1: + difficult_flag = True + obj_name = obj.find('name').text + if obj_name not in class_names: + continue + bndbox = obj.find('bndbox') + left = bndbox.find('xmin').text + top = bndbox.find('ymin').text + right = bndbox.find('xmax').text + bottom = bndbox.find('ymax').text + + if difficult_flag: + new_f.write("%s %s %s %s %s difficult\n" % (obj_name, left, top, right, bottom)) + else: + new_f.write("%s %s %s %s %s\n" % (obj_name, left, top, right, bottom)) + print("Get ground truth result done.") + + if map_mode == 0 or map_mode == 3: + print("Get map.") + get_map(MINOVERLAP, True, score_threhold = score_threhold, path = map_out_path) + print("Get map done.") + + if map_mode == 4: + print("Get map.") + get_coco_map(class_names = class_names, path = map_out_path) + print("Get map done.") diff --git a/model_data/2007_train.txt b/model_data/2007_train.txt new file mode 100644 index 0000000..958e509 --- /dev/null +++ b/model_data/2007_train.txt @@ -0,0 +1,639 @@ +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000002_000.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000005_000.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000005_002.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000005_004.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000005_005.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000006_000.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000007_000.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000008_001.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000011_001.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000011_002.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000011_003.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000011_004.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000011_008.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000013_000.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000013_008.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000013_014.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000013_015.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000013_016.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000013_017.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000013_019.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000014_000.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000015_000.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000016_000.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000017_001.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000017_002.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000018_000.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000022_000.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000023_001.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000023_003.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000029_000.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000032_003.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000032_005.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000032_009.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000032_010.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000032_019.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000032_020.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000032_022.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000032_025.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000032_046.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000032_048.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000032_049.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000032_051.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000034_001.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000035_000.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000037_000.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000038_004.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000039_000.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000039_001.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000039_002.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000039_003.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000040_000.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000040_002.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000041_003.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000041_004.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000042_000.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000042_002.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000042_004.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000042_008.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000046_000.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000047_000.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000047_001.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000047_002.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000047_005.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000047_007.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000048_000.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000049_001.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000049_002.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000050_000.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000050_001.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000050_002.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000050_003.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000052_000.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000052_001.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000054_001.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000054_002.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000054_003.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000054_005.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000054_008.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000055_000.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000072_000.png 178,283,242,325,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000147_001.png 308,299,359,389,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000149_006.png 301,230,426,275,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000181_061.png 104,284,211,411,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000211_010.png 176,267,227,351,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000211_019.png 170,211,420,414,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000211_041.png 112,208,368,423,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000344_003.png 86,115,135,226,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000377_004.png 193,209,445,393,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000398_003.png 184,194,426,382,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000457_004.png 401,140,448,193,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000468_017.png 74,259,130,282,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000468_033.png 102,251,160,277,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000583_008.png 146,227,251,257,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000643_002.png 409,269,457,307,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000661_000.png 149,218,423,432,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000732_005.png 213,232,420,404,0 306,55,393,107,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000740_000.png 150,192,389,385,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000744_006.png 73,254,144,314,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000756_001.png 144,162,445,377,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000808_002.png 279,192,392,275,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000830_000.png 287,244,428,302,3 154,199,203,259,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000845_000.png 185,183,412,396,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000865_006.png 419,312,458,332,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000902_001.png 175,317,219,359,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00001075_024.png 145,163,222,262,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00001153_004.png 153,126,202,346,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00001170_046.png 317,191,440,288,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00001248_038.png 283,47,410,131,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00001369_000.png 168,138,445,331,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00001373_009.png 206,234,480,484,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00001373_039.png 130,250,411,495,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00001534_005.png 198,171,443,361,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00001555_002.png 307,404,344,442,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00001558_016.png 42,209,156,417,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00001673_016.png 140,274,166,324,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00001836_041.png 313,192,329,205,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00001836_082.png 85,284,166,339,3 168,194,238,251,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00001946_029.png 144,198,181,275,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00002059_008.png 187,232,464,415,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00002106_000.png 162,181,188,207,2 318,214,383,291,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00002176_007.png 129,94,183,152,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00002290_001.png 130,358,161,382,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00002350_001.png 363,240,437,284,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00002533_002.png 140,13,244,280,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00002583_014.png 304,198,463,402,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00002763_031.png 201,177,439,371,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00003072_028.png 12,169,154,394,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00003148_004.png 298,252,326,342,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00003394_006.png 156,187,377,339,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00003400_003.png 304,28,424,142,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00003440_000.png 108,127,170,167,3 295,228,353,285,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00003787_003.png 144,235,228,294,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00003803_010.png 85,295,405,361,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00003945_004.png 34,235,187,319,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00003948_001.png 140,152,192,216,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00004296_000.png 365,376,425,423,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00004342_002.png 100,243,339,483,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00004342_050.png 92,180,160,391,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00004344_013.png 175,173,437,384,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00004344_014.png 164,156,398,381,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00004344_018.png 195,196,448,411,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00004344_022.png 188,254,420,441,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00004344_046.png 184,166,424,353,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00004381_021.png 194,180,421,343,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00004461_000.png 164,196,370,370,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00004533_014.png 225,153,449,349,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00004534_001.png 128,214,403,422,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00004547_003.png 449,345,476,374,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00004578_004.png 216,195,432,381,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00004630_001.png 142,219,398,391,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00004822_051.png 179,184,441,381,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00004893_085.png 192,171,424,366,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00004911_018.png 347,159,398,204,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00004968_003.png 320,342,453,395,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00004968_004.png 317,318,449,380,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00005066_005.png 165,243,419,418,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00005066_030.png 138,229,408,380,0 74,364,88,397,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00005089_002.png 324,287,455,381,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00005089_014.png 89,238,159,274,3 322,93,436,391,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00005089_040.png 309,49,434,389,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00005140_001.png 62,131,132,354,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00005532_000.png 191,131,429,338,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00005532_014.png 169,165,399,338,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00005532_016.png 162,208,369,378,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00005532_019.png 341,117,377,146,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00005827_000.png 196,153,422,389,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00005869_001.png 90,33,213,73,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00006736_000.png 385,236,407,259,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00006751_000.png 87,222,125,255,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00006821_002.png 60,42,251,380,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00006851_033.png 94,312,194,339,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00006851_034.png 125,308,205,369,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00006912_007.png 165,196,385,359,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00007037_000.png 144,243,394,436,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00007043_000.png 145,180,410,399,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00007120_009.png 124,274,192,318,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00007124_008.png 272,244,426,337,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00007471_003.png 270,53,437,294,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00007551_020.png 176,261,405,436,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00007557_026.png 65,353,176,415,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00007676_002.png 180,219,221,288,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00007710_000.png 312,219,420,263,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00007735_040.png 199,238,470,462,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00007882_001.png 75,321,176,350,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00008008_027.png 323,244,371,295,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00008339_010.png 174,173,437,366,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00008365_000.png 168,254,377,456,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00008386_000.png 355,178,385,202,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00008399_007.png 181,189,455,375,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00008522_032.png 157,104,374,301,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00008547_001.png 403,340,474,411,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00008554_009.png 303,247,418,331,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00008716_000.png 353,354,429,406,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00008814_010.png 97,203,432,236,3 97,206,430,259,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00008841_025.png 77,243,210,293,1 292,40,388,83,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00008841_044.png 73,23,235,106,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00009166_004.png 78,121,119,273,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00009218_015.png 96,109,207,405,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00009229_007.png 361,273,396,310,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00009256_005.png 152,196,208,254,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00009342_000.png 429,214,506,340,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00009403_006.png 374,208,412,234,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00009437_008.png 376,297,457,336,3 371,290,465,429,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00009608_024.png 197,201,379,372,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00009619_000.png 150,329,204,385,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00009669_003.png 78,324,209,434,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00009683_005.png 16,179,153,330,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00009705_000.png 169,154,440,339,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00009745_000.png 164,199,392,391,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00009779_001.png 355,247,437,281,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00009889_018.png 62,379,79,399,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00010007_168.png 166,360,194,420,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00010103_014.png 171,206,192,229,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00010125_004.png 73,150,106,194,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00010172_001.png 94,347,118,373,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00010277_000.png 431,346,467,402,1 148,155,418,294,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00010381_000.png 195,156,422,336,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00010478_012.png 106,293,229,394,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00010481_021.png 381,275,461,310,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00010575_002.png 116,302,233,381,3 347,295,471,383,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00010625_014.png 135,221,408,290,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00010767_008.png 101,95,184,166,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00010767_016.png 58,88,214,384,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00010770_000.png 122,185,225,219,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00010815_006.png 155,120,232,242,3 155,120,228,248,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00010828_039.png 294,234,400,348,3 351,223,470,342,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00010936_011.png 135,290,187,389,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00010936_016.png 87,82,203,151,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00010959_010.png 106,245,145,275,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00011151_004.png 165,103,353,286,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00011157_001.png 100,245,163,306,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00011237_094.png 64,114,144,369,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00011263_004.png 156,196,433,437,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00011269_019.png 366,209,475,303,1 286,133,351,204,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00011322_006.png 182,242,423,399,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00011402_007.png 102,240,349,412,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00011450_000.png 367,247,416,295,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00011463_002.png 189,272,410,444,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00011502_001.png 221,158,404,349,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00011576_000.png 103,291,126,312,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00011814_031.png 84,227,178,377,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00011827_003.png 284,235,437,297,3 376,74,440,366,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00011831_010.png 117,41,237,127,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00011832_002.png 372,175,407,212,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00011857_001.png 357,268,462,309,3 347,256,481,339,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00011925_072.png 188,127,269,329,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00011925_076.png 61,53,209,240,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00012045_009.png 358,280,408,339,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00012048_007.png 143,190,217,266,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00012094_040.png 288,55,361,108,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00012123_001.png 319,305,456,413,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00012174_000.png 84,79,178,349,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00012261_001.png 226,216,493,426,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00012270_005.png 371,117,428,173,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00012291_008.png 116,315,229,353,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00012374_000.png 327,128,368,168,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00012376_010.png 370,105,433,317,2 187,71,247,99,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00012376_011.png 385,240,422,279,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00012415_002.png 199,267,236,310,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00012505_007.png 127,84,246,350,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00012515_002.png 144,319,186,347,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00012622_016.png 381,83,422,147,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00012637_000.png 364,249,430,291,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00012670_000.png 213,210,428,368,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00012686_003.png 162,205,430,406,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00012741_004.png 221,217,444,421,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00012793_000.png 176,189,423,391,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00012834_008.png 55,206,154,242,3 2,179,139,390,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00012834_034.png 384,113,487,292,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00012834_122.png 38,47,167,484,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00012892_010.png 187,112,240,179,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00012973_005.png 61,31,201,412,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00012975_003.png 103,118,206,219,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00013031_005.png 9,209,89,406,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00013051_000.png 133,275,162,303,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00013111_069.png 237,201,309,331,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00013118_008.png 112,273,155,313,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00013125_000.png 84,203,415,412,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00013187_002.png 99,283,468,371,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00013249_031.png 191,239,393,368,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00013249_033.png 207,201,441,373,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00013285_026.png 89,249,108,383,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00013337_000.png 30,157,169,381,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00013346_015.png 186,211,416,472,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00013471_002.png 64,346,199,396,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00013508_001.png 161,296,250,341,3 143,258,188,305,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00013615_052.png 175,203,407,381,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00013635_002.png 181,165,412,350,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00013659_019.png 279,83,330,152,2 145,42,199,67,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00013674_000.png 105,159,134,188,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00013685_028.png 132,246,392,318,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00013750_016.png 98,91,195,228,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00013751_003.png 338,305,383,350,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00013807_009.png 157,61,218,82,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00013885_000.png 218,136,240,158,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00013911_000.png 59,241,101,292,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00013951_001.png 150,133,182,158,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00013977_005.png 390,392,452,482,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00013991_000.png 190,185,243,232,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00013992_006.png 140,263,204,303,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00013993_077.png 83,203,184,315,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00013993_083.png 68,108,196,457,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00014004_038.png 172,198,227,267,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00014014_013.png 324,191,409,276,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00014083_023.png 320,305,458,371,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00014095_003.png 148,248,234,295,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00014116_009.png 348,159,375,188,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00014177_010.png 94,194,158,310,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00014223_009.png 112,193,354,337,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00014253_010.png 18,123,141,165,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00014346_010.png 76,72,219,379,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00014398_031.png 94,231,189,395,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00014447_004.png 106,210,188,258,3 77,214,124,348,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00014551_010.png 293,205,372,295,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00014574_000.png 184,182,403,380,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00014607_007.png 100,330,421,437,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00014626_017.png 135,67,186,142,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00014626_035.png 75,85,183,478,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00014687_001.png 363,247,433,275,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00014706_018.png 185,223,397,368,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00014716_007.png 343,65,435,222,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00014731_028.png 386,132,443,203,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00014738_000.png 95,60,220,350,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00014795_002.png 144,270,218,337,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00014822_039.png 322,102,419,414,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00014870_004.png 57,306,169,353,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00015018_004.png 402,180,430,209,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00015141_002.png 398,301,423,334,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00015262_005.png 68,267,143,315,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00015300_000.png 85,197,159,271,3 107,237,146,264,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00015304_001.png 145,167,453,414,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00015400_001.png 149,175,389,380,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00015440_000.png 318,152,394,186,3 87,146,241,404,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00015530_147.png 93,33,195,340,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00015583_000.png 339,170,363,188,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00015649_000.png 320,149,363,186,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00015719_005.png 176,189,426,461,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00015770_010.png 181,235,407,370,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00015792_005.png 133,146,173,185,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00015794_000.png 34,324,58,346,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00015799_012.png 180,231,442,430,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00016009_008.png 286,288,363,412,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00016184_040.png 349,269,371,295,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00016191_017.png 341,296,441,378,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00016267_000.png 89,299,205,408,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00016291_002.png 308,338,471,377,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00016291_019.png 76,37,214,219,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00016414_000.png 192,215,478,439,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00016417_008.png 392,126,483,436,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00016429_015.png 73,46,263,331,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00016487_002.png 320,346,364,385,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00016490_011.png 91,261,175,293,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00016522_023.png 68,125,236,447,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00016568_010.png 345,200,399,262,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00016606_000.png 183,212,427,402,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00016837_002.png 70,329,206,376,3 70,352,199,383,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00016937_014.png 86,41,218,145,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00016964_011.png 364,276,411,306,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00016972_019.png 257,317,406,431,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00016972_025.png 344,262,404,360,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00016987_019.png 132,303,185,342,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00016987_022.png 41,244,190,305,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00016990_000.png 156,237,385,388,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00017028_000.png 72,289,111,319,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00017124_004.png 396,284,478,505,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00017137_016.png 85,73,228,311,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00017151_003.png 312,244,418,343,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00017178_007.png 172,150,395,339,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00017188_002.png 99,240,214,285,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00017199_005.png 386,341,422,378,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00017214_015.png 306,174,346,246,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00017243_010.png 38,301,82,342,2 10,145,73,428,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00017255_001.png 346,299,407,343,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00017257_001.png 213,112,372,287,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00017346_000.png 436,307,450,324,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00017500_002.png 382,307,435,338,3 55,248,154,384,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00017514_008.png 173,222,441,390,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00017524_028.png 125,217,374,432,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00017582_003.png 128,201,228,273,3 53,171,111,326,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00017611_002.png 129,224,210,299,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00017670_005.png 184,295,244,369,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00017714_006.png 406,254,468,418,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00017747_008.png 114,87,215,166,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00017877_001.png 330,284,430,323,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00017893_005.png 193,182,439,386,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00017972_026.png 331,123,377,226,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00018055_005.png 67,66,238,301,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00018055_045.png 44,154,67,204,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00018101_012.png 86,92,215,254,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00018102_001.png 53,294,73,321,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00018187_034.png 240,198,454,368,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00018233_057.png 173,192,379,366,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00018253_017.png 274,264,355,367,3 67,131,141,397,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00018253_059.png 92,202,194,346,3 89,191,190,339,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00018366_000.png 95,303,127,338,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00018366_029.png 280,235,405,268,1 297,64,399,162,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00018387_030.png 194,179,442,326,0 79,220,463,301,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00018427_004.png 47,315,135,352,3 385,175,477,449,1 266,164,319,214,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00018427_011.png 297,343,449,393,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00018496_006.png 40,354,476,464,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00018623_001.png 138,249,444,378,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00018657_003.png 361,224,446,371,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00018686_000.png 176,153,425,380,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00018693_004.png 150,217,424,417,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00018762_001.png 341,331,382,369,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00018762_002.png 301,308,430,349,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00018814_000.png 344,257,394,303,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00018865_008.png 104,66,177,112,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00018980_002.png 123,87,213,197,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00018984_000.png 412,154,449,212,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00019013_002.png 391,279,416,301,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00019058_004.png 175,229,221,299,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00019089_004.png 287,312,409,405,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00019124_045.png 166,260,229,294,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00019124_090.png 328,247,376,284,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00019124_104.png 59,69,206,239,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00019154_002.png 146,62,212,171,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00019187_000.png 198,191,430,384,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00019271_030.png 117,348,194,404,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00019271_064.png 132,254,429,388,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00019271_065.png 105,249,176,305,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00019373_036.png 333,203,419,259,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00019373_058.png 162,165,224,243,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00019399_010.png 391,301,458,425,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00019426_000.png 224,220,422,356,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00019495_004.png 369,282,443,302,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00019499_000.png 361,156,417,204,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00019706_002.png 288,44,364,104,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00019706_014.png 331,217,417,288,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00019766_023.png 367,208,463,366,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00019767_016.png 64,318,223,361,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00019861_010.png 134,178,402,377,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00019892_003.png 285,59,412,109,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00019924_020.png 315,314,410,504,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00020000_000.png 271,292,382,436,1 135,66,179,80,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00020065_008.png 409,158,453,228,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00020113_017.png 108,257,448,341,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00020124_003.png 87,290,209,341,3 180,142,234,210,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00020146_005.png 90,281,239,396,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00020213_078.png 92,335,211,385,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00020277_001.png 346,214,448,410,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00020318_007.png 357,271,467,329,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00020318_022.png 275,43,413,163,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00020332_000.png 114,311,152,343,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00020349_006.png 231,155,371,303,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00020393_001.png 118,113,156,148,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00020393_003.png 340,270,451,319,3 341,251,467,343,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00020405_041.png 55,143,154,332,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00020408_037.png 95,316,209,396,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00020438_011.png 164,180,407,420,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00020482_032.png 241,176,346,357,3 215,212,356,343,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00020564_000.png 387,179,444,299,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00020673_005.png 281,163,410,248,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00020751_003.png 373,77,497,406,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00020774_000.png 114,238,145,270,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00020810_003.png 79,357,209,414,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00020819_002.png 184,170,431,368,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00020857_008.png 95,285,376,387,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00020986_000.png 209,190,429,336,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00021007_000.png 172,234,224,285,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00021009_001.png 159,182,389,337,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00021024_022.png 359,261,441,356,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00021132_000.png 352,102,432,340,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00021179_011.png 62,86,239,420,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00021181_002.png 75,195,218,298,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00021201_010.png 357,95,435,254,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00021321_002.png 95,22,197,158,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00021374_000.png 373,188,433,241,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00021377_016.png 361,336,449,387,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00021381_013.png 281,289,422,334,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00021420_014.png 393,135,429,174,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00021420_020.png 83,345,129,390,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00021420_027.png 96,282,149,326,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00021443_000.png 143,186,415,393,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00021481_014.png 297,200,347,280,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00021700_010.png 293,41,394,124,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00021703_001.png 68,250,376,356,3 281,255,385,333,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00021711_014.png 275,334,451,395,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00021748_000.png 275,37,358,64,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00021772_016.png 85,96,131,137,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00021782_028.png 66,49,177,270,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00021796_000.png 93,205,181,295,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00021840_016.png 71,268,192,317,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00021845_001.png 153,186,347,368,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00021860_002.png 50,262,121,367,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00021896_003.png 119,68,232,176,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00021967_000.png 91,46,273,339,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00022141_023.png 322,48,428,113,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00022155_008.png 342,138,420,177,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00022192_003.png 87,110,218,262,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00022215_012.png 161,176,418,385,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00022237_002.png 134,236,198,287,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00022369_013.png 83,199,170,293,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00022416_004.png 409,234,481,334,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00022416_018.png 178,211,462,388,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00022416_049.png 394,213,436,278,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00022572_063.png 35,258,176,418,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00022572_087.png 10,122,117,353,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00022611_001.png 374,84,459,272,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00022706_001.png 173,157,398,351,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00022707_003.png 377,284,463,320,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00022883_002.png 57,255,423,309,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00022977_000.png 120,321,153,353,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00023058_004.png 19,281,177,382,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00023068_003.png 175,373,193,390,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00023075_033.png 119,267,156,300,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00023078_000.png 370,343,396,367,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00023078_003.png 47,330,112,387,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00023093_007.png 345,206,458,377,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00023093_009.png 216,176,410,312,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00023156_001.png 114,115,192,173,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00023162_025.png 96,39,228,168,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00023168_000.png 107,339,154,400,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00023176_010.png 140,252,478,331,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00023176_017.png 161,127,428,340,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00023325_019.png 172,140,410,315,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00025221_001.png 269,291,441,451,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00025228_005.png 295,229,443,349,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00025228_007.png 59,235,130,386,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00025252_032.png 378,340,478,462,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00025252_040.png 300,101,431,310,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00025252_053.png 7,320,175,400,1 325,319,397,451,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00025252_054.png 329,360,401,472,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00025270_000.png 134,260,184,294,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00025368_014.png 147,342,205,384,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00025368_018.png 75,61,235,197,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00025368_033.png 379,139,478,236,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00025521_003.png 36,260,89,390,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00025529_018.png 362,83,454,385,1 351,194,406,258,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00025662_006.png 413,319,439,348,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00025664_002.png 33,72,171,387,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00025686_000.png 298,335,464,387,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00025732_004.png 109,195,378,403,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00025747_000.png 167,185,415,369,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00025769_001.png 350,286,402,318,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00025787_027.png 302,49,490,418,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00025787_039.png 43,49,134,334,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00025787_050.png 305,90,333,215,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00025954_025.png 98,46,235,171,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00026087_000.png 169,231,378,428,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00026136_002.png 275,277,414,332,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00026196_001.png 196,245,238,285,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00026221_001.png 99,364,151,422,2 68,75,228,328,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00026285_000.png 452,412,476,436,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00026319_000.png 334,326,394,365,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00026338_003.png 176,172,395,330,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00026392_005.png 281,55,440,183,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00026398_000.png 182,132,206,156,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00026451_068.png 147,321,209,368,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00026586_009.png 98,93,183,158,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00026695_000.png 170,98,229,164,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00026753_008.png 299,53,450,393,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00026769_010.png 396,208,423,235,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00026810_001.png 383,354,442,381,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00026886_002.png 112,98,229,437,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00026886_004.png 117,258,224,326,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00026889_000.png 118,185,415,412,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00026911_005.png 173,153,216,201,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00026920_000.png 171,223,232,249,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00026983_001.png 276,95,310,131,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00027028_017.png 75,260,461,330,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00027094_003.png 384,280,461,356,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00027103_001.png 336,118,438,253,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00027213_044.png 285,83,389,489,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00027278_007.png 76,23,232,162,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00027357_014.png 31,52,162,375,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00027441_002.png 60,67,169,348,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00027470_006.png 199,250,247,287,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00027474_005.png 138,203,188,235,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00027577_003.png 47,105,115,238,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00027631_000.png 418,356,473,427,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00027652_003.png 37,101,110,358,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00027685_003.png 170,161,404,352,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00027697_001.png 354,192,451,221,3 79,244,234,350,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00027797_000.png 215,204,462,365,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00027833_022.png 150,152,445,373,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00027866_002.png 72,308,373,416,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00027875_005.png 348,156,481,357,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00027927_009.png 369,195,395,226,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00027937_004.png 83,263,138,314,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00028012_001.png 280,231,443,331,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00028018_000.png 176,226,404,440,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00028027_000.png 287,123,346,195,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00028173_016.png 282,250,338,319,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00028330_003.png 104,219,177,317,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00028383_002.png 389,385,433,414,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00028452_001.png 314,205,442,279,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00028509_007.png 24,169,128,451,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00028509_026.png 155,200,408,371,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00028518_021.png 92,339,185,392,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00028607_000.png 202,151,450,372,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00028620_000.png 399,334,476,376,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00028625_000.png 169,225,229,273,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00028628_008.png 303,49,409,210,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00028628_015.png 84,76,170,281,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00028628_020.png 313,169,365,211,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00028640_008.png 381,271,448,405,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00028698_001.png 425,202,455,286,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00028861_009.png 290,23,436,173,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00028873_009.png 166,140,364,319,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00028876_027.png 268,56,384,128,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00028876_035.png 95,170,127,206,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00028924_009.png 365,112,427,242,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00028974_016.png 384,290,467,389,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00029075_013.png 328,67,453,182,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00029088_023.png 152,147,225,208,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00029105_015.png 114,214,174,275,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00029200_006.png 66,254,144,311,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00029259_027.png 393,193,450,339,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00029391_000.png 157,185,400,373,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00029464_006.png 45,142,131,239,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00029464_015.png 99,176,407,338,3 359,305,455,331,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00029469_011.png 362,215,448,441,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00029502_006.png 110,294,236,384,3 74,226,202,423,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00029532_005.png 64,49,206,280,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00029579_014.png 50,260,69,322,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00029588_004.png 20,99,153,381,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00029631_006.png 59,254,145,307,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00029647_002.png 167,203,415,396,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00029807_003.png 82,41,260,192,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00029817_009.png 110,158,188,267,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00029843_001.png 128,224,193,279,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00029894_000.png 357,254,472,363,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00029906_000.png 165,224,381,382,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00029909_003.png 83,84,200,129,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00029940_007.png 106,191,178,282,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00030039_008.png 13,60,145,373,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00030106_008.png 112,258,227,393,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00030111_007.png 296,48,375,174,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00030162_029.png 163,190,197,227,2 310,83,412,163,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00030206_013.png 156,169,389,386,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00030260_004.png 184,275,212,386,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00030323_015.png 307,211,420,357,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00030394_001.png 311,91,344,120,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00030408_013.png 359,302,418,374,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00030434_000.png 333,322,371,442,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00030606_006.png 275,15,393,96,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00030636_004.png 305,220,385,295,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00030674_000.png 316,125,430,231,3 diff --git a/model_data/2007_val.txt b/model_data/2007_val.txt new file mode 100644 index 0000000..6b0e8f1 --- /dev/null +++ b/model_data/2007_val.txt @@ -0,0 +1,72 @@ +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000005_001.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000005_003.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000013_038.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000032_008.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00000042_005.png +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00001320_003.png 104,272,140,287,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00001437_012.png 41,69,218,377,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00002176_005.png 130,44,223,122,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00002395_007.png 115,224,153,303,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00002435_005.png 181,240,382,389,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00002578_000.png 398,163,437,193,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00002856_009.png 108,318,217,369,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00004342_020.png 158,193,395,409,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00006621_004.png 153,200,261,308,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00007034_016.png 383,288,428,376,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00009368_006.png 184,140,218,194,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00010071_008.png 101,340,137,431,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00011291_003.png 346,73,441,175,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00011322_002.png 173,178,426,420,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00011355_011.png 250,26,464,253,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00012045_019.png 211,246,239,274,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00012094_047.png 333,71,400,147,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00012299_002.png 97,287,427,327,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00012829_004.png 84,269,168,316,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00013310_059.png 96,104,190,388,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00013673_001.png 75,217,115,253,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00013685_000.png 129,218,185,243,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00013911_021.png 49,121,399,390,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00014198_000.png 338,256,387,352,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00014626_028.png 166,177,385,377,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00014976_003.png 303,340,424,380,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00015090_006.png 319,285,417,374,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00015425_012.png 192,144,415,322,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00015895_017.png 359,177,438,438,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00016403_003.png 446,438,501,471,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00016564_000.png 192,195,418,348,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00016587_069.png 377,203,412,232,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00016624_000.png 276,173,460,339,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00017138_037.png 59,143,154,384,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00017403_006.png 297,288,393,440,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00017448_000.png 177,241,416,437,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00017511_006.png 169,211,368,378,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00018187_029.png 171,210,427,441,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00018253_054.png 91,154,214,407,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00018412_001.png 106,308,392,353,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00018496_007.png 286,62,397,109,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00019313_000.png 374,307,423,361,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00019917_004.png 306,388,430,412,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00020274_021.png 79,218,130,253,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00020408_058.png 80,36,224,232,4 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00020429_020.png 83,295,174,338,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00021364_001.png 167,225,415,378,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00021409_001.png 156,172,402,365,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00021862_004.png 303,279,363,347,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00022098_006.png 247,288,382,365,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00022726_002.png 344,200,407,260,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00023116_005.png 320,219,410,298,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00025969_000.png 182,142,401,332,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00026132_016.png 350,241,390,278,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00026261_001.png 98,140,217,260,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00027479_013.png 386,282,479,317,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00027556_007.png 86,322,155,380,3 35,98,120,216,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00027817_001.png 380,195,478,378,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00028265_007.png 399,303,442,353,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00028873_004.png 160,160,380,328,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00028876_060.png 381,194,419,237,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00029464_003.png 121,187,224,251,3 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00029808_003.png 160,205,384,402,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00030279_000.png 144,210,418,438,0 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00030413_003.png 127,118,172,157,2 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00030634_000.png 70,305,229,354,1 +E:\project\python\rabat-illness-yolov5\database\Train/JPEGImages/00030635_001.png 288,245,408,356,3 diff --git a/model_data/coco_classes.txt b/model_data/coco_classes.txt index cc9b5e3..8a1218a 100644 --- a/model_data/coco_classes.txt +++ b/model_data/coco_classes.txt @@ -1,2 +1,5 @@ -0 -1 \ No newline at end of file +1 +2 +3 +4 +5 diff --git a/model_data/voc_classes.txt b/model_data/voc_classes.txt index cc9b5e3..8a1218a 100644 --- a/model_data/voc_classes.txt +++ b/model_data/voc_classes.txt @@ -1,2 +1,5 @@ -0 -1 \ No newline at end of file +1 +2 +3 +4 +5 diff --git a/model_data/yolov5_s_v6.1.pth b/model_data/yolov5_s_v6.1.pth index 63bfedd..ed2b67b 100644 Binary files a/model_data/yolov5_s_v6.1.pth and b/model_data/yolov5_s_v6.1.pth differ diff --git a/nets/CSPdarknet.py b/nets/CSPdarknet.py new file mode 100644 index 0000000..7f7299e --- /dev/null +++ b/nets/CSPdarknet.py @@ -0,0 +1,177 @@ +import torch +import torch.nn as nn + + +class SiLU(nn.Module): + @staticmethod + def forward(x): + return x * torch.sigmoid(x) + +def autopad(k, p=None): + if p is None: + p = k // 2 if isinstance(k, int) else [x // 2 for x in k] + return p + +class Focus(nn.Module): + def __init__(self, c1, c2, k=1, s=1, p=None, g=1, act=True): # ch_in, ch_out, kernel, stride, padding, groups + super(Focus, self).__init__() + self.conv = Conv(c1 * 4, c2, k, s, p, g, act) + + def forward(self, x): + # 320, 320, 12 => 320, 320, 64 + return self.conv( + # 640, 640, 3 => 320, 320, 12 + torch.cat( + [ + x[..., ::2, ::2], + x[..., 1::2, ::2], + x[..., ::2, 1::2], + x[..., 1::2, 1::2] + ], 1 + ) + ) + +class Conv(nn.Module): + def __init__(self, c1, c2, k=1, s=1, p=None, g=1, act=True): + super(Conv, self).__init__() + self.conv = nn.Conv2d(c1, c2, k, s, autopad(k, p), groups=g, bias=False) + self.bn = nn.BatchNorm2d(c2, eps=0.001, momentum=0.03) + self.act = SiLU() if act is True else (act if isinstance(act, nn.Module) else nn.Identity()) + + def forward(self, x): + return self.act(self.bn(self.conv(x))) + + def fuseforward(self, x): + return self.act(self.conv(x)) + +class Bottleneck(nn.Module): + # Standard bottleneck + def __init__(self, c1, c2, shortcut=True, g=1, e=0.5): # ch_in, ch_out, shortcut, groups, expansion + super(Bottleneck, self).__init__() + c_ = int(c2 * e) # hidden channels + self.cv1 = Conv(c1, c_, 1, 1) + self.cv2 = Conv(c_, c2, 3, 1, g=g) + self.add = shortcut and c1 == c2 + + def forward(self, x): + return x + self.cv2(self.cv1(x)) if self.add else self.cv2(self.cv1(x)) + +class C3(nn.Module): + # CSP Bottleneck with 3 convolutions + def __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5): # ch_in, ch_out, number, shortcut, groups, expansion + super(C3, self).__init__() + c_ = int(c2 * e) # hidden channels + self.cv1 = Conv(c1, c_, 1, 1) + self.cv2 = Conv(c1, c_, 1, 1) + self.cv3 = Conv(2 * c_, c2, 1) # act=FReLU(c2) + self.m = nn.Sequential(*[Bottleneck(c_, c_, shortcut, g, e=1.0) for _ in range(n)]) + # self.m = nn.Sequential(*[CrossConv(c_, c_, 3, 1, g, 1.0, shortcut) for _ in range(n)]) + + def forward(self, x): + return self.cv3(torch.cat( + ( + self.m(self.cv1(x)), + self.cv2(x) + ) + , dim=1)) + +class SPP(nn.Module): + # Spatial pyramid pooling layer used in YOLOv3-SPP + def __init__(self, c1, c2, k=(5, 9, 13)): + super(SPP, self).__init__() + c_ = c1 // 2 # hidden channels + self.cv1 = Conv(c1, c_, 1, 1) + self.cv2 = Conv(c_ * (len(k) + 1), c2, 1, 1) + self.m = nn.ModuleList([nn.MaxPool2d(kernel_size=x, stride=1, padding=x // 2) for x in k]) + + def forward(self, x): + x = self.cv1(x) + return self.cv2(torch.cat([x] + [m(x) for m in self.m], 1)) + +class CSPDarknet(nn.Module): + def __init__(self, base_channels, base_depth, phi, pretrained): + super().__init__() + #-----------------------------------------------# + # 输入图片是640, 640, 3 + # 初始的基本通道base_channels是64 + #-----------------------------------------------# + + #-----------------------------------------------# + # 利用focus网络结构进行特征提取 + # 640, 640, 3 -> 320, 320, 12 -> 320, 320, 64 + #-----------------------------------------------# + self.stem = Focus(3, base_channels, k=3) + + #-----------------------------------------------# + # 完成卷积之后，320, 320, 64 -> 160, 160, 128 + # 完成CSPlayer之后，160, 160, 128 -> 160, 160, 128 + #-----------------------------------------------# + self.dark2 = nn.Sequential( + # 320, 320, 64 -> 160, 160, 128 + Conv(base_channels, base_channels * 2, 3, 2), + # 160, 160, 128 -> 160, 160, 128 + C3(base_channels * 2, base_channels * 2, base_depth), + ) + + #-----------------------------------------------# + # 完成卷积之后，160, 160, 128 -> 80, 80, 256 + # 完成CSPlayer之后，80, 80, 256 -> 80, 80, 256 + # 在这里引出有效特征层80, 80, 256 + # 进行加强特征提取网络FPN的构建 + #-----------------------------------------------# + self.dark3 = nn.Sequential( + Conv(base_channels * 2, base_channels * 4, 3, 2), + C3(base_channels * 4, base_channels * 4, base_depth * 3), + ) + + #-----------------------------------------------# + # 完成卷积之后，80, 80, 256 -> 40, 40, 512 + # 完成CSPlayer之后，40, 40, 512 -> 40, 40, 512 + # 在这里引出有效特征层40, 40, 512 + # 进行加强特征提取网络FPN的构建 + #-----------------------------------------------# + self.dark4 = nn.Sequential( + Conv(base_channels * 4, base_channels * 8, 3, 2), + C3(base_channels * 8, base_channels * 8, base_depth * 3), + ) + + #-----------------------------------------------# + # 完成卷积之后，40, 40, 512 -> 20, 20, 1024 + # 完成SPP之后，20, 20, 1024 -> 20, 20, 1024 + # 完成CSPlayer之后，20, 20, 1024 -> 20, 20, 1024 + #-----------------------------------------------# + self.dark5 = nn.Sequential( + Conv(base_channels * 8, base_channels * 16, 3, 2), + SPP(base_channels * 16, base_channels * 16), + C3(base_channels * 16, base_channels * 16, base_depth, shortcut=False), + ) + if pretrained: + url = { + 's' : 'https://github.com/bubbliiiing/yolov5-pytorch/releases/download/v1.0/cspdarknet_s_backbone.pth', + 'm' : 'https://github.com/bubbliiiing/yolov5-pytorch/releases/download/v1.0/cspdarknet_m_backbone.pth', + 'l' : 'https://github.com/bubbliiiing/yolov5-pytorch/releases/download/v1.0/cspdarknet_l_backbone.pth', + 'x' : 'https://github.com/bubbliiiing/yolov5-pytorch/releases/download/v1.0/cspdarknet_x_backbone.pth', + }[phi] + checkpoint = torch.hub.load_state_dict_from_url(url=url, map_location="cpu", model_dir="./model_data") + self.load_state_dict(checkpoint, strict=False) + print("Load weights from ", url.split('/')[-1]) + + def forward(self, x): + x = self.stem(x) + x = self.dark2(x) + #-----------------------------------------------# + # dark3的输出为80, 80, 256，是一个有效特征层 + #-----------------------------------------------# + x = self.dark3(x) + feat1 = x + #-----------------------------------------------# + # dark4的输出为40, 40, 512，是一个有效特征层 + #-----------------------------------------------# + x = self.dark4(x) + feat2 = x + #-----------------------------------------------# + # dark5的输出为20, 20, 1024，是一个有效特征层 + #-----------------------------------------------# + x = self.dark5(x) + feat3 = x + return feat1, feat2, feat3 diff --git a/nets/ConvNext.py b/nets/ConvNext.py new file mode 100644 index 0000000..1687a3d --- /dev/null +++ b/nets/ConvNext.py @@ -0,0 +1,249 @@ +# Copyright (c) Meta Platforms, Inc. and affiliates. + +# All rights reserved. + +# This source code is licensed under the license found in the +# LICENSE file in the root directory of this source tree. + + +import math + +import numpy as np +import torch +import torch.nn as nn +import torch.nn.functional as F + + +def drop_path(x, drop_prob: float = 0., training: bool = False, scale_by_keep: bool = True): + if drop_prob == 0. or not training: + return x + keep_prob = 1 - drop_prob + shape = (x.shape[0],) + (1,) * (x.ndim - 1) + random_tensor = x.new_empty(shape).bernoulli_(keep_prob) + if keep_prob > 0.0 and scale_by_keep: + random_tensor.div_(keep_prob) + return x * random_tensor + +class DropPath(nn.Module): + """ + Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks). + """ + def __init__(self, drop_prob=None, scale_by_keep=True): + super(DropPath, self).__init__() + self.drop_prob = drop_prob + self.scale_by_keep = scale_by_keep + + def forward(self, x): + return drop_path(x, self.drop_prob, self.training, self.scale_by_keep) + +def trunc_normal_(tensor, mean=0., std=1., a=-2., b=2.): + def _no_grad_trunc_normal_(tensor, mean, std, a, b): + def norm_cdf(x): + return (1. + math.erf(x / math.sqrt(2.))) / 2. + + with torch.no_grad(): + l = norm_cdf((a - mean) / std) + u = norm_cdf((b - mean) / std) + + tensor.uniform_(2 * l - 1, 2 * u - 1) + tensor.erfinv_() + + tensor.mul_(std * math.sqrt(2.)) + tensor.add_(mean) + + tensor.clamp_(min=a, max=b) + return tensor + return _no_grad_trunc_normal_(tensor, mean, std, a, b) + +#--------------------------------------# +# Gelu激活函数的实现 +# 利用近似的数学公式 +#--------------------------------------# +class GELU(nn.Module): + def __init__(self): + super(GELU, self).__init__() + + def forward(self, x): + return 0.5 * x * (1 + torch.tanh(np.sqrt(2 / np.pi) * (x + 0.044715 * torch.pow(x,3)))) + +#---------------------------------------------------------------------------------# +# LayerNorm 支持两种形式channels_last (default) or channels_first. +# channels_last 对应具有形状的输入(batch_size, height, width, channels) +# channels_first 对应具有形状的输入(batch_size, channels, height, width). +#---------------------------------------------------------------------------------# +class LayerNorm(nn.Module): + def __init__(self, normalized_shape, eps=1e-6, data_format="channels_last"): + super().__init__() + self.weight = nn.Parameter(torch.ones(normalized_shape)) + self.bias = nn.Parameter(torch.zeros(normalized_shape)) + self.eps = eps + self.data_format = data_format + if self.data_format not in ["channels_last", "channels_first"]: + raise NotImplementedError + self.normalized_shape = (normalized_shape, ) + + def forward(self, x): + if self.data_format == "channels_last": + return F.layer_norm(x, self.normalized_shape, self.weight, self.bias, self.eps) + elif self.data_format == "channels_first": + u = x.mean(1, keepdim=True) + s = (x - u).pow(2).mean(1, keepdim=True) + x = (x - u) / torch.sqrt(s + self.eps) + x = self.weight[:, None, None] * x + self.bias[:, None, None] + return x + +#--------------------------------------------------------------------------------------------------------------# +# ConvNeXt Block有两种等效的实现: +# (1) DwConv -> LayerNorm (channels_first) -> 1x1 Conv -> GELU -> 1x1 Conv; all in (N, C, H, W) +# (2) DwConv -> Permute to (N, H, W, C); LayerNorm (channels_last) -> Linear -> GELU -> Linear; Permute back +# 代码中使用（2），因为这个在PyTorch中稍微快一点 +#--------------------------------------------------------------------------------------------------------------# +class Block(nn.Module): + def __init__(self, dim, drop_path=0., layer_scale_init_value=1e-6): + super().__init__() + #--------------------------# + # 7x7的逐层卷积 + #--------------------------# + self.dwconv = nn.Conv2d(dim, dim, kernel_size=7, padding=3, groups=dim) + self.norm = LayerNorm(dim, eps=1e-6) + #--------------------------# + # 利用全连接层代替1x1卷积 + #--------------------------# + self.pwconv1 = nn.Linear(dim, 4 * dim) + self.act = GELU() + #--------------------------# + # 利用全连接层代替1x1卷积 + #--------------------------# + self.pwconv2 = nn.Linear(4 * dim, dim) + #--------------------------# + # 加入缩放系数 + #--------------------------# + self.gamma = nn.Parameter(layer_scale_init_value * torch.ones((dim)), requires_grad=True) if layer_scale_init_value > 0 else None + #--------------------------# + # 加入Drop_path正则化 + #--------------------------# + self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity() + + def forward(self, x): + input = x + #--------------------------# + # 7x7的逐层卷积 + #--------------------------# + x = self.dwconv(x) + x = x.permute(0, 2, 3, 1) # (N, C, H, W) -> (N, H, W, C) + x = self.norm(x) + #--------------------------# + # 利用全连接层代替1x1卷积 + #--------------------------# + x = self.pwconv1(x) + x = self.act(x) + #--------------------------# + # 利用全连接层代替1x1卷积 + #--------------------------# + x = self.pwconv2(x) + #--------------------------# + # 加入缩放系数 + #--------------------------# + if self.gamma is not None: + x = self.gamma * x + x = x.permute(0, 3, 1, 2) # (N, H, W, C) -> (N, C, H, W) + #--------------------------# + # 加入Drop_path正则化 + #--------------------------# + x = input + self.drop_path(x) + return x + +#-----------------------------------------------------# +# ConvNeXt +# A PyTorch impl of : `A ConvNet for the 2020s` +# https://arxiv.org/pdf/2201.03545.pdf +#-----------------------------------------------------# +class ConvNeXt(nn.Module): + def __init__( + self, in_chans=3, num_classes=1000, depths=[3, 3, 9, 3], dims=[96, 192, 384, 768], + drop_path_rate=0., layer_scale_init_value=1e-6, head_init_scale=1., **kwargs + ): + super().__init__() + + self.downsample_layers = nn.ModuleList() + #--------------------------------------------------# + # bs, 3, 224, 224 -> bs, 96, 56, 56 + #--------------------------------------------------# + stem = nn.Sequential( + nn.Conv2d(in_chans, dims[0], kernel_size=4, stride=4), + LayerNorm(dims[0], eps=1e-6, data_format="channels_first") + ) + self.downsample_layers.append(stem) + + #--------------------------------------------------# + # 定义三次下采样的过程 + # 利用步长为2x2，卷积核大小为2x2的卷积进行下采样 + #--------------------------------------------------# + for i in range(3): + downsample_layer = nn.Sequential( + LayerNorm(dims[i], eps=1e-6, data_format="channels_first"), + nn.Conv2d(dims[i], dims[i+1], kernel_size=2, stride=2), + ) + self.downsample_layers.append(downsample_layer) + + #--------------------------------------------------# + # 根据深度的不同，定义不同的drop率 + #--------------------------------------------------# + self.stages = nn.ModuleList() + dp_rates = [x.item() for x in torch.linspace(0, drop_path_rate, sum(depths))] + cur = 0 + #--------------------------------------------------# + # 整个ConvNeXt除了Stem外，存在四个Stage + # 每个Stage里面是多个ConvNeXt Block的堆叠。 + #--------------------------------------------------# + for i in range(4): + stage = nn.Sequential( + *[Block(dim=dims[i], drop_path=dp_rates[cur + j], layer_scale_init_value=layer_scale_init_value) for j in range(depths[i])] + ) + self.stages.append(stage) + cur += depths[i] + + self.apply(self._init_weights) + + def _init_weights(self, m): + if isinstance(m, (nn.Conv2d, nn.Linear)): + trunc_normal_(m.weight, std=.02) + nn.init.constant_(m.bias, 0) + + def forward(self, x): + outs = [] + for i in range(4): + x = self.downsample_layers[i](x) + x = self.stages[i](x) + if i != 0: + outs.append(x) + return outs + +model_urls = { + "convnext_tiny_1k" : "https://github.com/bubbliiiing/yolov5-pytorch/releases/download/v1.0/convnext_tiny_1k_224_ema_no_jit.pth", + "convnext_small_1k" : "https://github.com/bubbliiiing/yolov5-pytorch/releases/download/v1.0/convnext_small_1k_224_ema_no_jit.pth", +} + +#------------------------------------------------------# +# Tiny约等于Cspdarknet-L的尺寸 +#------------------------------------------------------# +def ConvNeXt_Tiny(pretrained=False, **kwargs): + model = ConvNeXt(depths=[3, 3, 9, 3], dims=[96, 192, 384, 768], **kwargs) + if pretrained: + url = model_urls['convnext_tiny_1k'] + checkpoint = torch.hub.load_state_dict_from_url(url=url, map_location="cpu", model_dir="./model_data") + model.load_state_dict(checkpoint, strict=False) + print("Load weights from ", url.split('/')[-1]) + return model + +#------------------------------------------------------# +# Tiny约等于Cspdarknet-X的尺寸 +#------------------------------------------------------# +def ConvNeXt_Small(pretrained=False, **kwargs): + model = ConvNeXt(depths=[3, 3, 27, 3], dims=[96, 192, 384, 768], **kwargs) + if pretrained: + url = model_urls['convnext_small_1k'] + checkpoint = torch.hub.load_state_dict_from_url(url=url, map_location="cpu", model_dir="./model_data") + model.load_state_dict(checkpoint, strict=False) + print("Load weights from ", url.split('/')[-1]) + return model diff --git a/nets/Swin_transformer.py b/nets/Swin_transformer.py new file mode 100644 index 0000000..8a302e6 --- /dev/null +++ b/nets/Swin_transformer.py @@ -0,0 +1,638 @@ +# -------------------------------------------------------- +# Swin Transformer +# Copyright (c) 2021 Microsoft +# Licensed under The MIT License [see LICENSE for details] +# Written by Ze Liu +# -------------------------------------------------------- +import math + +import numpy as np +import torch +import torch.nn as nn +import torch.nn.functional as F +import torch.utils.checkpoint as checkpoint + + +def _make_divisible(v, divisor, min_value=None): + if min_value is None: + min_value = divisor + new_v = max(min_value, int(v + divisor / 2) // divisor * divisor) + if new_v < 0.9 * v: + new_v += divisor + return new_v + +def trunc_normal_(tensor, mean=0., std=1., a=-2., b=2.): + def _no_grad_trunc_normal_(tensor, mean, std, a, b): + def norm_cdf(x): + return (1. + math.erf(x / math.sqrt(2.))) / 2. + + with torch.no_grad(): + l = norm_cdf((a - mean) / std) + u = norm_cdf((b - mean) / std) + + tensor.uniform_(2 * l - 1, 2 * u - 1) + tensor.erfinv_() + + tensor.mul_(std * math.sqrt(2.)) + tensor.add_(mean) + + tensor.clamp_(min=a, max=b) + return tensor + return _no_grad_trunc_normal_(tensor, mean, std, a, b) + +#--------------------------------------# +# Gelu激活函数的实现 +# 利用近似的数学公式 +#--------------------------------------# +class GELU(nn.Module): + def __init__(self): + super(GELU, self).__init__() + + def forward(self, x): + return 0.5 * x * (1 + torch.tanh(np.sqrt(2 / np.pi) * (x + 0.044715 * torch.pow(x,3)))) + +#-------------------------------------------------------# +# 对输入进来的图片进行高和宽的压缩 +# 并且进行通道的扩张。 +#-------------------------------------------------------# +class PatchEmbed(nn.Module): + def __init__(self, img_size=[224, 224], patch_size=4, in_chans=3, embed_dim=96, norm_layer=None): + super().__init__() + # [224, 224] + self.img_size = img_size + # [4, 4] + self.patch_size = [patch_size, patch_size] + # [56, 56] + self.patches_resolution = [self.img_size[0] // self.patch_size[0], self.img_size[1] // self.patch_size[1]] + + # 3136 + self.num_patches = self.patches_resolution[0] * self.patches_resolution[1] + # 3 + self.in_chans = in_chans + # 96 + self.embed_dim = embed_dim + + #-------------------------------------------------------# + # bs, 224, 224, 3 -> bs, 56, 56, 96 + #-------------------------------------------------------# + self.proj = nn.Conv2d(in_chans, embed_dim, kernel_size=patch_size, stride=patch_size) + if norm_layer is not None: + self.norm = norm_layer(embed_dim) + else: + self.norm = None + + def forward(self, x): + B, C, H, W = x.shape + # FIXME look at relaxing size constraints + assert H == self.img_size[0] and W == self.img_size[1], \ + f"Input image size ({H}*{W}) doesn't match model ({self.img_size[0]} * {self.img_size[1]})." + #-------------------------------------------------------# + # bs, 224, 224, 3 -> bs, 56, 56, 96 -> bs, 3136, 96 + #-------------------------------------------------------# + x = self.proj(x).flatten(2).transpose(1, 2) + if self.norm is not None: + x = self.norm(x) + return x + +def window_partition(x, window_size): + B, H, W, C = x.shape + #------------------------------------------------------------------# + # bs, 56, 56, 96 -> bs, 8, 7, 8, 7, 96 -> bs * 64, 7, 7, 96 + #------------------------------------------------------------------# + x = x.view(B, H // window_size, window_size, W // window_size, window_size, C) + windows = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(-1, window_size, window_size, C) + return windows + +def window_reverse(windows, window_size, H, W): + #------------------------------------------------------------------# + # bs * 64, 7, 7, 96 -> bs, 8, 8, 7, 7, 96 -> bs, 56, 56, 96 + #------------------------------------------------------------------# + B = int(windows.shape[0] / (H * W / window_size / window_size)) + x = windows.view(B, H // window_size, W // window_size, window_size, window_size, -1) + x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, H, W, -1) + return x + + +class WindowAttention(nn.Module): + def __init__(self, dim, window_size, num_heads, qkv_bias=True, qk_scale=None, attn_drop=0., proj_drop=0.): + super().__init__() + self.dim = dim + self.window_size = window_size # Wh, Ww + self.num_heads = num_heads + head_dim = dim // num_heads + self.scale = qk_scale or head_dim ** -0.5 + + #--------------------------------------------------------------------------# + # 相对坐标矩阵，用于表示每个窗口内，其它点相对于自己的坐标 + # 由于相对坐标取值范围为-6 ~ +6。中间共13个值，因此需要13 * 13 + # 13 * 13, num_heads + #--------------------------------------------------------------------------# + self.relative_position_bias_table = nn.Parameter( + torch.zeros((2 * window_size[0] - 1) * (2 * window_size[1] - 1), num_heads) + ) + + #--------------------------------------------------------------------------# + # 该部分用于获取7x7的矩阵内部，其它特征点相对于自身相对坐标 + #--------------------------------------------------------------------------# + coords_h = torch.arange(self.window_size[0]) + coords_w = torch.arange(self.window_size[1]) + coords = torch.stack(torch.meshgrid([coords_h, coords_w])) # 2, Wh, Ww + coords_flatten = torch.flatten(coords, 1) # 2, Wh*Ww + relative_coords = coords_flatten[:, :, None] - coords_flatten[:, None, :] # 2, Wh*Ww, Wh*Ww + relative_coords = relative_coords.permute(1, 2, 0).contiguous() # Wh*Ww, Wh*Ww, 2 + relative_coords[:, :, 0] += self.window_size[0] - 1 # shift to start from 0 + relative_coords[:, :, 1] += self.window_size[1] - 1 + relative_coords[:, :, 0] *= 2 * self.window_size[1] - 1 + relative_position_index = relative_coords.sum(-1) # Wh*Ww, Wh*Ww + self.register_buffer("relative_position_index", relative_position_index) + + #--------------------------------------------------------------------------# + # 乘积获得q、k、v，用于计算多头注意力机制 + #--------------------------------------------------------------------------# + self.qkv = nn.Linear(dim, dim * 3, bias=qkv_bias) + self.attn_drop = nn.Dropout(attn_drop) + self.proj = nn.Linear(dim, dim) + self.proj_drop = nn.Dropout(proj_drop) + + trunc_normal_(self.relative_position_bias_table, std=.02) + self.softmax = nn.Softmax(dim=-1) + + def forward(self, x, mask=None): + B_, N, C = x.shape + #--------------------------------------------------------------------------# + # bs * 64, 49, 96 -> bs * 64, 49, 96 * 3 -> + # bs * 64, 49, 3, num_heads, 32 -> 3, bs * 64, num_head, 49, 32 + #--------------------------------------------------------------------------# + qkv = self.qkv(x).reshape(B_, N, 3, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4) + #--------------------------------------------------------------------------# + # bs * 64, num_head, 49, 32 + #--------------------------------------------------------------------------# + q, k, v = qkv[0], qkv[1], qkv[2] + + #--------------------------------------------------------------------------# + # bs * 64, num_head, 49, 49 + #--------------------------------------------------------------------------# + q = q * self.scale + attn = (q @ k.transpose(-2, -1)) + + #--------------------------------------------------------------------------# + # 这一步是根据已经求得的注意力，加上相对坐标的偏执量 + # 形成最后的注意力 + #--------------------------------------------------------------------------# + relative_position_bias = self.relative_position_bias_table[self.relative_position_index.view(-1)].view( + self.window_size[0] * self.window_size[1], self.window_size[0] * self.window_size[1], -1) # Wh*Ww,Wh*Ww,nH + relative_position_bias = relative_position_bias.permute(2, 0, 1).contiguous() + attn = attn + relative_position_bias.unsqueeze(0) + + #--------------------------------------------------------------------------# + # 加上mask，保证分区。 + # bs * 64, num_head, 49, 49 + #--------------------------------------------------------------------------# + if mask is not None: + nW = mask.shape[0] + attn = attn.view(B_ // nW, nW, self.num_heads, N, N) + mask.unsqueeze(1).unsqueeze(0) + attn = attn.view(-1, self.num_heads, N, N) + attn = self.softmax(attn) + else: + attn = self.softmax(attn) + + attn = self.attn_drop(attn) + + #---------------------------------------------------------------------------------------# + # bs * 64, num_head, 49, 49 @ bs * 64, num_head, 49, 32 -> bs * 64, num_head, 49, 32 + # + # bs * 64, num_head, 49, 32 -> bs * 64, 49, 96 + #---------------------------------------------------------------------------------------# + x = (attn @ v).transpose(1, 2).reshape(B_, N, C) + x = self.proj(x) + x = self.proj_drop(x) + return x + + +def drop_path(x, drop_prob: float = 0., training: bool = False, scale_by_keep: bool = True): + """ + Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks). + This is the same as the DropConnect impl I created for EfficientNet, etc networks, however, + the original name is misleading as 'Drop Connect' is a different form of dropout in a separate paper... + See discussion: https://github.com/tensorflow/tpu/issues/494#issuecomment-532968956 ... I've opted for + changing the layer and argument names to 'drop path' rather than mix DropConnect as a layer name and use + 'survival rate' as the argument. + """ + if drop_prob == 0. or not training: + return x + keep_prob = 1 - drop_prob + shape = (x.shape[0],) + (1,) * (x.ndim - 1) # work with diff dim tensors, not just 2D ConvNets + random_tensor = x.new_empty(shape).bernoulli_(keep_prob) + if keep_prob > 0.0 and scale_by_keep: + random_tensor.div_(keep_prob) + return x * random_tensor + + +class DropPath(nn.Module): + """ + Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks). + """ + def __init__(self, drop_prob=None, scale_by_keep=True): + super(DropPath, self).__init__() + self.drop_prob = drop_prob + self.scale_by_keep = scale_by_keep + + def forward(self, x): + return drop_path(x, self.drop_prob, self.training, self.scale_by_keep) + + +#-------------------------------------------------------# +# 两次全连接 +#-------------------------------------------------------# +class Mlp(nn.Module): + def __init__(self, in_features, hidden_features=None, out_features=None, act_layer=GELU, drop=0.): + super().__init__() + out_features = out_features or in_features + hidden_features = hidden_features or in_features + self.fc1 = nn.Linear(in_features, hidden_features) + self.act = act_layer() + self.fc2 = nn.Linear(hidden_features, out_features) + self.drop = nn.Dropout(drop) + + def forward(self, x): + x = self.fc1(x) + x = self.act(x) + x = self.drop(x) + x = self.fc2(x) + x = self.drop(x) + return x + +#-------------------------------------------------------# +# 每个阶段重复的基础模块 +# 在这其中会使用WindowAttention进行特征提取 +#-------------------------------------------------------# +class SwinTransformerBlock(nn.Module): + def __init__(self, dim, input_resolution, num_heads, window_size=7, shift_size=0, + mlp_ratio=4., qkv_bias=True, qk_scale=None, drop=0., attn_drop=0., drop_path=0., + act_layer=GELU, norm_layer=nn.LayerNorm): + super().__init__() + self.dim = dim + self.input_resolution = input_resolution + self.num_heads = num_heads + self.window_size = window_size + self.shift_size = shift_size + + self.mlp_ratio = mlp_ratio + if min(self.input_resolution) <= self.window_size: + self.shift_size = 0 + self.window_size = min(self.input_resolution) + assert 0 <= self.shift_size < self.window_size, "shift_size must in 0-window_size" + + self.norm1 = norm_layer(dim) + self.attn = WindowAttention( + dim, + window_size = [self.window_size, self.window_size], + num_heads = num_heads, + qkv_bias = qkv_bias, + qk_scale = qk_scale, + attn_drop = attn_drop, + proj_drop = drop + ) + + self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity() + self.norm2 = norm_layer(dim) + mlp_hidden_dim = int(dim * mlp_ratio) + self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim, act_layer=act_layer, drop=drop) + + if self.shift_size > 0: + #----------------------------------------------------------------# + # 由于进行特征提取时，会对输入的特征层进行的平移 + # 如： + # [ [ + # [1, 2, 3], [5, 6, 4], + # [4, 5, 6], --> [8, 9, 7], + # [7, 8, 9], [1, 2, 3], + # ] ] + # 这一步的作用就是使得平移后的区域块只计算自己部分的注意力机制 + #----------------------------------------------------------------# + H, W = self.input_resolution + _H, _W = _make_divisible(H, self.window_size), _make_divisible(W, self.window_size), + img_mask = torch.zeros((1, _H, _W, 1)) # 1 H W 1 + h_slices = (slice(0, -self.window_size), + slice(-self.window_size, -self.shift_size), + slice(-self.shift_size, None)) + w_slices = (slice(0, -self.window_size), + slice(-self.window_size, -self.shift_size), + slice(-self.shift_size, None)) + cnt = 0 + for h in h_slices: + for w in w_slices: + img_mask[:, h, w, :] = cnt + cnt += 1 + + mask_windows = window_partition(img_mask, self.window_size) # nW, window_size, window_size, 1 + mask_windows = mask_windows.view(-1, self.window_size * self.window_size) + attn_mask = mask_windows.unsqueeze(1) - mask_windows.unsqueeze(2) + attn_mask = attn_mask.masked_fill(attn_mask != 0, float(-100.0)).masked_fill(attn_mask == 0, float(0.0)) + self.attn_mask = attn_mask.cpu().numpy() + else: + self.attn_mask = None + + def forward(self, x): + H, W = self.input_resolution + B, L, C = x.shape + assert L == H * W, "input feature has wrong size" + #-----------------------------------------------# + # bs, 3136, 96 -> bs, 56, 56, 96 + #-----------------------------------------------# + shortcut = x + x = self.norm1(x) + x = x.view(B, H, W, C) + + _H, _W = _make_divisible(H, self.window_size), _make_divisible(W, self.window_size), + x = x.permute(0, 3, 1, 2) + x = F.interpolate(x, [_H, _W], mode='bicubic', align_corners=False).permute(0, 2, 3, 1) + + #-----------------------------------------------# + # 进行特征层的平移 + #-----------------------------------------------# + if self.shift_size > 0: + shifted_x = torch.roll(x, shifts=(-self.shift_size, -self.shift_size), dims=(1, 2)) + else: + shifted_x = x + #------------------------------------------------------------------------------------------# + # bs, 56, 56, 96 -> bs * 64, 7, 7, 96 -> bs * 64, 49, 96 + #------------------------------------------------------------------------------------------# + x_windows = window_partition(shifted_x, self.window_size) # num_windows * B, window_size, window_size, C + x_windows = x_windows.view(-1, self.window_size * self.window_size, C) # nW*B, window_size*window_size, C + + #-----------------------------------------------# + # bs * 64, 49, 97 -> bs * 64, 49, 97 + #-----------------------------------------------# + if type(self.attn_mask) != type(None): + attn_mask = torch.tensor(self.attn_mask).cuda() if x.is_cuda else torch.tensor(self.attn_mask) + else: + attn_mask = None + attn_windows = self.attn(x_windows, mask=attn_mask) # nW*B, window_size*window_size, C + #-----------------------------------------------# + # bs * 64, 49, 97 -> bs, 56, 56, 96 + #-----------------------------------------------# + attn_windows = attn_windows.view(-1, self.window_size, self.window_size, C) + shifted_x = window_reverse(attn_windows, self.window_size, _H, _W) # B H' W' C + + #-----------------------------------------------# + # 将特征层平移回来 + #-----------------------------------------------# + if self.shift_size > 0: + x = torch.roll(shifted_x, shifts=(self.shift_size, self.shift_size), dims=(1, 2)) + else: + x = shifted_x + + x = x.permute(0, 3, 1, 2) + x = F.interpolate(x, [H, W], mode='bicubic', align_corners=False).permute(0, 2, 3, 1) + #-----------------------------------------------# + # bs, 3136, 96 + #-----------------------------------------------# + x = x.view(B, H * W, C) + #-----------------------------------------------# + # FFN + # bs, 3136, 96 + #-----------------------------------------------# + x = shortcut + self.drop_path(x) + x = x + self.drop_path(self.mlp(self.norm2(x))) + + return x + +#-------------------------------------------------------# +# 对输入进来的特征层进行高和宽的压缩 +# 进行跨特征点的特征提取，提取完成后进行堆叠。 +#-------------------------------------------------------# +class PatchMerging(nn.Module): + def __init__(self, input_resolution, dim, norm_layer=nn.LayerNorm): + super().__init__() + self.input_resolution = input_resolution + self.dim = dim + + self.norm = norm_layer(4 * dim) + self.reduction = nn.Linear(4 * dim, 2 * dim, bias=False) + + def forward(self, x): + H, W = self.input_resolution + B, L, C = x.shape + assert L == H * W, "input feature has wrong size" + assert H % 2 == 0 and W % 2 == 0, f"x size ({H}*{W}) are not even." + + #-------------------------------------------------------# + # bs, 3136, 96 -> bs, 56, 56, 96 + #-------------------------------------------------------# + x = x.view(B, H, W, C) + + #-------------------------------------------------------# + # x0 ~ x3 bs, 56, 56, 96 -> bs, 28, 28, 96 + #-------------------------------------------------------# + x0 = x[:, 0::2, 0::2, :] # B H/2 W/2 C + x1 = x[:, 1::2, 0::2, :] # B H/2 W/2 C + x2 = x[:, 0::2, 1::2, :] # B H/2 W/2 C + x3 = x[:, 1::2, 1::2, :] # B H/2 W/2 C + + #-------------------------------------------------------# + # 4 X bs, 28, 28, 96 -> bs, 28, 28, 384 + #-------------------------------------------------------# + x = torch.cat([x0, x1, x2, x3], -1) # B H/2 W/2 4*C + #-------------------------------------------------------# + # bs, 28, 28, 384 -> bs, 784, 384 + #-------------------------------------------------------# + x = x.view(B, -1, 4 * C) # B H/2*W/2 4*C + + #-------------------------------------------------------# + # bs, 784, 384 -> bs, 784, 192 + #-------------------------------------------------------# + x = self.norm(x) + x = self.reduction(x) + return x + + +#-------------------------------------------------------# +# Swin-Transformer的基础模块。 +# 使用窗口多头注意力机制进行特征提取。 +# 使用PatchMerging进行高和宽的压缩。 +#-------------------------------------------------------# +class BasicLayer(nn.Module): + def __init__(self, dim, input_resolution, depth, num_heads, window_size, + mlp_ratio=4., qkv_bias=True, qk_scale=None, drop=0., attn_drop=0., + drop_path=0., norm_layer=nn.LayerNorm, downsample=None, use_checkpoint=False): + super().__init__() + #-------------------------------------------------------# + # 四个阶段对应不同的dim + # [96, 192, 384, 768] + #-------------------------------------------------------# + self.dim = dim + #-------------------------------------------------------# + # 四个阶段对应不同的输入分辨率 + # [[56, 56], [28, 28], [14, 14], [7, 7]] + #-------------------------------------------------------# + self.input_resolution = input_resolution + #-------------------------------------------------------# + # 四个阶段对应不同的多头注意力机制重复次数 + # [2, 2, 6, 2] + #-------------------------------------------------------# + self.depth = depth + self.use_checkpoint = use_checkpoint + + #-------------------------------------------------------# + # 根据depth的次数利用窗口多头注意力机制进行特征提取。 + #-------------------------------------------------------# + self.blocks = nn.ModuleList( + [ + SwinTransformerBlock( + dim = dim, + input_resolution = input_resolution, + num_heads = num_heads, + window_size = window_size, + shift_size = 0 if (i % 2 == 0) else window_size // 2, + mlp_ratio = mlp_ratio, + qkv_bias = qkv_bias, + qk_scale = qk_scale, + drop = drop, + attn_drop = attn_drop, + drop_path = drop_path[i] if isinstance(drop_path, list) else drop_path, + norm_layer = norm_layer + ) + for i in range(depth) + ] + ) + + if downsample is not None: + #-------------------------------------------------------# + # 判断是否要进行下采样，即：高宽压缩 + #-------------------------------------------------------# + self.downsample = downsample(input_resolution, dim=dim, norm_layer=norm_layer) + else: + self.downsample = None + + def forward(self, x): + for blk in self.blocks: + if self.use_checkpoint: + x_ = checkpoint.checkpoint(blk, x) + else: + x_ = blk(x) + if self.downsample is not None: + x = self.downsample(x_) + else: + x = x_ + return x_, x + +class SwinTransformer(nn.Module): + def __init__(self, img_size=[640, 640], patch_size=4, in_chans=3, num_classes=1000, + embed_dim=96, depths=[2, 2, 6, 2], num_heads=[3, 6, 12, 24], + window_size=7, mlp_ratio=4., qkv_bias=True, qk_scale=None, + drop_rate=0., attn_drop_rate=0., drop_path_rate=0.1, + norm_layer=nn.LayerNorm, ape=False, patch_norm=True, + use_checkpoint=False, **kwargs): + super().__init__() + self.num_classes = num_classes + self.num_layers = len(depths) + self.embed_dim = embed_dim + self.ape = ape + self.patch_norm = patch_norm + self.num_features = int(embed_dim * 2 ** (self.num_layers - 1)) + self.mlp_ratio = mlp_ratio + + #--------------------------------------------------# + # bs, 224, 224, 3 -> bs, 3136, 96 + #--------------------------------------------------# + self.patch_embed = PatchEmbed( + img_size = img_size, + patch_size = patch_size, + in_chans = in_chans, + embed_dim = embed_dim, + norm_layer = norm_layer if self.patch_norm else None + ) + + #--------------------------------------------------# + # PatchEmbed之后的图像序列长度 3136 + # PatchEmbed之后的图像对应的分辨率 [56, 56] + #--------------------------------------------------# + num_patches = self.patch_embed.num_patches + patches_resolution = self.patch_embed.patches_resolution + self.patches_resolution = patches_resolution + + if self.ape: + self.absolute_pos_embed = nn.Parameter(torch.zeros(1, num_patches, embed_dim)) + trunc_normal_(self.absolute_pos_embed, std=.02) + + self.pos_drop = nn.Dropout(p=drop_rate) + + #--------------------------------------------------# + # stochastic depth + #--------------------------------------------------# + dpr = [x.item() for x in torch.linspace(0, drop_path_rate, sum(depths))] # stochastic depth decay rule + + #---------------------------------------------------------------# + # 构建swin-transform的每个阶段 + # bs, 3136, 96 -> bs, 784, 192 -> bs, 196, 384 -> bs, 49, 768 + #---------------------------------------------------------------# + self.layers = nn.ModuleList() + for i_layer in range(self.num_layers): + layer = BasicLayer( + dim = int(embed_dim * 2 ** i_layer), + input_resolution = (patches_resolution[0] // (2 ** i_layer), patches_resolution[1] // (2 ** i_layer)), + depth = depths[i_layer], + num_heads = num_heads[i_layer], + window_size = window_size, + mlp_ratio = self.mlp_ratio, + qkv_bias = qkv_bias, + qk_scale = qk_scale, + drop = drop_rate, + attn_drop = attn_drop_rate, + drop_path = dpr[sum(depths[:i_layer]):sum(depths[:i_layer + 1])], + norm_layer = norm_layer, + downsample = PatchMerging if (i_layer < self.num_layers - 1) else None, + use_checkpoint = use_checkpoint + ) + self.layers.append(layer) + + self.apply(self._init_weights) + + def _init_weights(self, m): + if isinstance(m, nn.Linear): + trunc_normal_(m.weight, std=.02) + if isinstance(m, nn.Linear) and m.bias is not None: + nn.init.constant_(m.bias, 0) + elif isinstance(m, nn.LayerNorm): + nn.init.constant_(m.bias, 0) + nn.init.constant_(m.weight, 1.0) + + @torch.jit.ignore + def no_weight_decay(self): + return {'absolute_pos_embed'} + + @torch.jit.ignore + def no_weight_decay_keywords(self): + return {'relative_position_bias_table'} + + def forward(self, x): + x = self.patch_embed(x) + if self.ape: + x = x + self.absolute_pos_embed + x = self.pos_drop(x) + + inverval_outs = [] + for i, layer in enumerate(self.layers): + x_, x = layer(x) + if i != 0: + inverval_outs.append(x_) + + outs = [] + for i, layer in enumerate(inverval_outs): + H, W = (self.patches_resolution[0] // (2 ** (i + 1)), self.patches_resolution[1] // (2 ** (i + 1))) + B, L, C = layer.shape + layer = layer.view([B, H, W, C]).permute([0, 3, 1, 2]) + outs.append(layer) + + return outs + +def Swin_transformer_Tiny(pretrained = False, input_shape = [640, 640], **kwargs): + model = SwinTransformer(input_shape, depths=[2, 2, 6, 2], **kwargs) + if pretrained: + url = "https://github.com/bubbliiiing/yolov5-pytorch/releases/download/v1.0/swin_tiny_patch4_window7.pth" + checkpoint = torch.hub.load_state_dict_from_url(url=url, map_location="cpu", model_dir="./model_data") + model.load_state_dict(checkpoint, strict=False) + print("Load weights from ", url.split('/')[-1]) + + return model \ No newline at end of file diff --git a/nets/__init__.py b/nets/__init__.py new file mode 100644 index 0000000..4287ca8 --- /dev/null +++ b/nets/__init__.py @@ -0,0 +1 @@ +# \ No newline at end of file diff --git a/nets/yolo.py b/nets/yolo.py new file mode 100644 index 0000000..f41331f --- /dev/null +++ b/nets/yolo.py @@ -0,0 +1,132 @@ +import torch +import torch.nn as nn + +from nets.ConvNext import ConvNeXt_Small, ConvNeXt_Tiny +from nets.CSPdarknet import C3, Conv, CSPDarknet +from nets.Swin_transformer import Swin_transformer_Tiny + + +#---------------------------------------------------# +# yolo_body +#---------------------------------------------------# +class YoloBody(nn.Module): + def __init__(self, anchors_mask, num_classes, phi, backbone='cspdarknet', pretrained=False, input_shape=[640, 640]): + super(YoloBody, self).__init__() + depth_dict = {'s' : 0.33, 'm' : 0.67, 'l' : 1.00, 'x' : 1.33,} + width_dict = {'s' : 0.50, 'm' : 0.75, 'l' : 1.00, 'x' : 1.25,} + dep_mul, wid_mul = depth_dict[phi], width_dict[phi] + + base_channels = int(wid_mul * 64) # 64 + base_depth = max(round(dep_mul * 3), 1) # 3 + #-----------------------------------------------# + # 输入图片是640, 640, 3 + # 初始的基本通道是64 + #-----------------------------------------------# + self.backbone_name = backbone + if backbone == "cspdarknet": + #---------------------------------------------------# + # 生成CSPdarknet53的主干模型 + # 获得三个有效特征层，他们的shape分别是： + # 80,80,256 + # 40,40,512 + # 20,20,1024 + #---------------------------------------------------# + self.backbone = CSPDarknet(base_channels, base_depth, phi, pretrained) + else: + #---------------------------------------------------# + # 如果输入不为cspdarknet，则调整通道数 + # 使其符合YoloV5的格式 + #---------------------------------------------------# + self.backbone = { + 'convnext_tiny' : ConvNeXt_Tiny, + 'convnext_small' : ConvNeXt_Small, + 'swin_transfomer_tiny' : Swin_transformer_Tiny, + }[backbone](pretrained=pretrained, input_shape=input_shape) + in_channels = { + 'convnext_tiny' : [192, 384, 768], + 'convnext_small' : [192, 384, 768], + 'swin_transfomer_tiny' : [192, 384, 768], + }[backbone] + feat1_c, feat2_c, feat3_c = in_channels + self.conv_1x1_feat1 = Conv(feat1_c, base_channels * 4, 1, 1) + self.conv_1x1_feat2 = Conv(feat2_c, base_channels * 8, 1, 1) + self.conv_1x1_feat3 = Conv(feat3_c, base_channels * 16, 1, 1) + + self.upsample = nn.Upsample(scale_factor=2, mode="nearest") + + self.conv_for_feat3 = Conv(base_channels * 16, base_channels * 8, 1, 1) + self.conv3_for_upsample1 = C3(base_channels * 16, base_channels * 8, base_depth, shortcut=False) + + self.conv_for_feat2 = Conv(base_channels * 8, base_channels * 4, 1, 1) + self.conv3_for_upsample2 = C3(base_channels * 8, base_channels * 4, base_depth, shortcut=False) + + self.down_sample1 = Conv(base_channels * 4, base_channels * 4, 3, 2) + self.conv3_for_downsample1 = C3(base_channels * 8, base_channels * 8, base_depth, shortcut=False) + + self.down_sample2 = Conv(base_channels * 8, base_channels * 8, 3, 2) + self.conv3_for_downsample2 = C3(base_channels * 16, base_channels * 16, base_depth, shortcut=False) + + # 80, 80, 256 => 80, 80, 3 * (5 + num_classes) => 80, 80, 3 * (4 + 1 + num_classes) + self.yolo_head_P3 = nn.Conv2d(base_channels * 4, len(anchors_mask[2]) * (5 + num_classes), 1) + # 40, 40, 512 => 40, 40, 3 * (5 + num_classes) => 40, 40, 3 * (4 + 1 + num_classes) + self.yolo_head_P4 = nn.Conv2d(base_channels * 8, len(anchors_mask[1]) * (5 + num_classes), 1) + # 20, 20, 1024 => 20, 20, 3 * (5 + num_classes) => 20, 20, 3 * (4 + 1 + num_classes) + self.yolo_head_P5 = nn.Conv2d(base_channels * 16, len(anchors_mask[0]) * (5 + num_classes), 1) + + def forward(self, x): + # backbone + feat1, feat2, feat3 = self.backbone(x) + if self.backbone_name != "cspdarknet": + feat1 = self.conv_1x1_feat1(feat1) + feat2 = self.conv_1x1_feat2(feat2) + feat3 = self.conv_1x1_feat3(feat3) + + # 20, 20, 1024 -> 20, 20, 512 + P5 = self.conv_for_feat3(feat3) + # 20, 20, 512 -> 40, 40, 512 + P5_upsample = self.upsample(P5) + # 40, 40, 512 -> 40, 40, 1024 + P4 = torch.cat([P5_upsample, feat2], 1) + # 40, 40, 1024 -> 40, 40, 512 + P4 = self.conv3_for_upsample1(P4) + + # 40, 40, 512 -> 40, 40, 256 + P4 = self.conv_for_feat2(P4) + # 40, 40, 256 -> 80, 80, 256 + P4_upsample = self.upsample(P4) + # 80, 80, 256 cat 80, 80, 256 -> 80, 80, 512 + P3 = torch.cat([P4_upsample, feat1], 1) + # 80, 80, 512 -> 80, 80, 256 + P3 = self.conv3_for_upsample2(P3) + + # 80, 80, 256 -> 40, 40, 256 + P3_downsample = self.down_sample1(P3) + # 40, 40, 256 cat 40, 40, 256 -> 40, 40, 512 + P4 = torch.cat([P3_downsample, P4], 1) + # 40, 40, 512 -> 40, 40, 512 + P4 = self.conv3_for_downsample1(P4) + + # 40, 40, 512 -> 20, 20, 512 + P4_downsample = self.down_sample2(P4) + # 20, 20, 512 cat 20, 20, 512 -> 20, 20, 1024 + P5 = torch.cat([P4_downsample, P5], 1) + # 20, 20, 1024 -> 20, 20, 1024 + P5 = self.conv3_for_downsample2(P5) + + #---------------------------------------------------# + # 第三个特征层 + # y3=(batch_size,75,80,80) + #---------------------------------------------------# + out2 = self.yolo_head_P3(P3) + #---------------------------------------------------# + # 第二个特征层 + # y2=(batch_size,75,40,40) + #---------------------------------------------------# + out1 = self.yolo_head_P4(P4) + #---------------------------------------------------# + # 第一个特征层 + # y1=(batch_size,75,20,20) + #---------------------------------------------------# + out0 = self.yolo_head_P5(P5) + return out0, out1, out2 + diff --git a/nets/yolo_training.py b/nets/yolo_training.py new file mode 100644 index 0000000..240e515 --- /dev/null +++ b/nets/yolo_training.py @@ -0,0 +1,465 @@ +import math +from copy import deepcopy +from functools import partial + +import numpy as np +import torch +import torch.nn as nn + + +class YOLOLoss(nn.Module): + def __init__(self, anchors, num_classes, input_shape, cuda, anchors_mask = [[6,7,8], [3,4,5], [0,1,2]], label_smoothing = 0): + super(YOLOLoss, self).__init__() + #-----------------------------------------------------------# + # 20x20的特征层对应的anchor是[116,90],[156,198],[373,326] + # 40x40的特征层对应的anchor是[30,61],[62,45],[59,119] + # 80x80的特征层对应的anchor是[10,13],[16,30],[33,23] + #-----------------------------------------------------------# + self.anchors = anchors + self.num_classes = num_classes + self.bbox_attrs = 5 + num_classes + self.input_shape = input_shape + self.anchors_mask = anchors_mask + self.label_smoothing = label_smoothing + + self.threshold = 4 + + self.balance = [0.4, 1.0, 4] + self.box_ratio = 0.05 + self.obj_ratio = 1 * (input_shape[0] * input_shape[1]) / (640 ** 2) + self.cls_ratio = 0.5 * (num_classes / 80) + self.cuda = cuda + + def clip_by_tensor(self, t, t_min, t_max): + t = t.float() + result = (t >= t_min).float() * t + (t < t_min).float() * t_min + result = (result <= t_max).float() * result + (result > t_max).float() * t_max + return result + + def MSELoss(self, pred, target): + return torch.pow(pred - target, 2) + + def BCELoss(self, pred, target): + epsilon = 1e-7 + pred = self.clip_by_tensor(pred, epsilon, 1.0 - epsilon) + output = - target * torch.log(pred) - (1.0 - target) * torch.log(1.0 - pred) + return output + + def box_giou(self, b1, b2): + """ + 输入为： + ---------- + b1: tensor, shape=(batch, feat_w, feat_h, anchor_num, 4), xywh + b2: tensor, shape=(batch, feat_w, feat_h, anchor_num, 4), xywh + + 返回为： + ------- + giou: tensor, shape=(batch, feat_w, feat_h, anchor_num, 1) + """ + #----------------------------------------------------# + # 求出预测框左上角右下角 + #----------------------------------------------------# + b1_xy = b1[..., :2] + b1_wh = b1[..., 2:4] + b1_wh_half = b1_wh/2. + b1_mins = b1_xy - b1_wh_half + b1_maxes = b1_xy + b1_wh_half + #----------------------------------------------------# + # 求出真实框左上角右下角 + #----------------------------------------------------# + b2_xy = b2[..., :2] + b2_wh = b2[..., 2:4] + b2_wh_half = b2_wh/2. + b2_mins = b2_xy - b2_wh_half + b2_maxes = b2_xy + b2_wh_half + + #----------------------------------------------------# + # 求真实框和预测框所有的iou + #----------------------------------------------------# + intersect_mins = torch.max(b1_mins, b2_mins) + intersect_maxes = torch.min(b1_maxes, b2_maxes) + intersect_wh = torch.max(intersect_maxes - intersect_mins, torch.zeros_like(intersect_maxes)) + intersect_area = intersect_wh[..., 0] * intersect_wh[..., 1] + b1_area = b1_wh[..., 0] * b1_wh[..., 1] + b2_area = b2_wh[..., 0] * b2_wh[..., 1] + union_area = b1_area + b2_area - intersect_area + iou = intersect_area / union_area + + #----------------------------------------------------# + # 找到包裹两个框的最小框的左上角和右下角 + #----------------------------------------------------# + enclose_mins = torch.min(b1_mins, b2_mins) + enclose_maxes = torch.max(b1_maxes, b2_maxes) + enclose_wh = torch.max(enclose_maxes - enclose_mins, torch.zeros_like(intersect_maxes)) + #----------------------------------------------------# + # 计算对角线距离 + #----------------------------------------------------# + enclose_area = enclose_wh[..., 0] * enclose_wh[..., 1] + giou = iou - (enclose_area - union_area) / enclose_area + + return giou + + #---------------------------------------------------# + # 平滑标签 + #---------------------------------------------------# + def smooth_labels(self, y_true, label_smoothing, num_classes): + return y_true * (1.0 - label_smoothing) + label_smoothing / num_classes + + def forward(self, l, input, targets=None, y_true=None): + #----------------------------------------------------# + # l 代表使用的是第几个有效特征层 + # input的shape为 bs, 3*(5+num_classes), 20, 20 + # bs, 3*(5+num_classes), 40, 40 + # bs, 3*(5+num_classes), 80, 80 + # targets 真实框的标签情况 [batch_size, num_gt, 5] + #----------------------------------------------------# + #--------------------------------# + # 获得图片数量，特征层的高和宽 + # 20, 20 + #--------------------------------# + bs = input.size(0) + in_h = input.size(2) + in_w = input.size(3) + #-----------------------------------------------------------------------# + # 计算步长 + # 每一个特征点对应原来的图片上多少个像素点 + # [640, 640] 高的步长为640 / 20 = 32，宽的步长为640 / 20 = 32 + # 如果特征层为20x20的话，一个特征点就对应原来的图片上的32个像素点 + # 如果特征层为40x40的话，一个特征点就对应原来的图片上的16个像素点 + # 如果特征层为80x80的话，一个特征点就对应原来的图片上的8个像素点 + # stride_h = stride_w = 32、16、8 + #-----------------------------------------------------------------------# + stride_h = self.input_shape[0] / in_h + stride_w = self.input_shape[1] / in_w + #-------------------------------------------------# + # 此时获得的scaled_anchors大小是相对于特征层的 + #-------------------------------------------------# + scaled_anchors = [(a_w / stride_w, a_h / stride_h) for a_w, a_h in self.anchors] + #-----------------------------------------------# + # 输入的input一共有三个，他们的shape分别是 + # bs, 3 * (5+num_classes), 20, 20 => bs, 3, 5 + num_classes, 20, 20 => batch_size, 3, 20, 20, 5 + num_classes + + # batch_size, 3, 20, 20, 5 + num_classes + # batch_size, 3, 40, 40, 5 + num_classes + # batch_size, 3, 80, 80, 5 + num_classes + #-----------------------------------------------# + prediction = input.view(bs, len(self.anchors_mask[l]), self.bbox_attrs, in_h, in_w).permute(0, 1, 3, 4, 2).contiguous() + + #-----------------------------------------------# + # 先验框的中心位置的调整参数 + #-----------------------------------------------# + x = torch.sigmoid(prediction[..., 0]) + y = torch.sigmoid(prediction[..., 1]) + #-----------------------------------------------# + # 先验框的宽高调整参数 + #-----------------------------------------------# + w = torch.sigmoid(prediction[..., 2]) + h = torch.sigmoid(prediction[..., 3]) + #-----------------------------------------------# + # 获得置信度，是否有物体 + #-----------------------------------------------# + conf = torch.sigmoid(prediction[..., 4]) + #-----------------------------------------------# + # 种类置信度 + #-----------------------------------------------# + pred_cls = torch.sigmoid(prediction[..., 5:]) + #-----------------------------------------------# + # self.get_target已经合并到dataloader中 + # 原因是在这里执行过慢，会大大延长训练时间 + #-----------------------------------------------# + # y_true, noobj_mask = self.get_target(l, targets, scaled_anchors, in_h, in_w) + + #---------------------------------------------------------------# + # 将预测结果进行解码，判断预测结果和真实值的重合程度 + # 如果重合程度过大则忽略，因为这些特征点属于预测比较准确的特征点 + # 作为负样本不合适 + #----------------------------------------------------------------# + pred_boxes = self.get_pred_boxes(l, x, y, h, w, targets, scaled_anchors, in_h, in_w) + + if self.cuda: + y_true = y_true.type_as(x) + + loss = 0 + n = torch.sum(y_true[..., 4] == 1) + if n != 0: + #---------------------------------------------------------------# + # 计算预测结果和真实结果的giou，计算对应有真实框的先验框的giou损失 + # loss_cls计算对应有真实框的先验框的分类损失 + #----------------------------------------------------------------# + giou = self.box_giou(pred_boxes, y_true[..., :4]).type_as(x) + loss_loc = torch.mean((1 - giou)[y_true[..., 4] == 1]) + loss_cls = torch.mean(self.BCELoss(pred_cls[y_true[..., 4] == 1], self.smooth_labels(y_true[..., 5:][y_true[..., 4] == 1], self.label_smoothing, self.num_classes))) + loss += loss_loc * self.box_ratio + loss_cls * self.cls_ratio + #-----------------------------------------------------------# + # 计算置信度的loss + # 也就意味着先验框对应的预测框预测的更准确 + # 它才是用来预测这个物体的。 + #-----------------------------------------------------------# + tobj = torch.where(y_true[..., 4] == 1, giou.detach().clamp(0), torch.zeros_like(y_true[..., 4])) + else: + tobj = torch.zeros_like(y_true[..., 4]) + loss_conf = torch.mean(self.BCELoss(conf, tobj)) + + loss += loss_conf * self.balance[l] * self.obj_ratio + # if n != 0: + # print(loss_loc * self.box_ratio, loss_cls * self.cls_ratio, loss_conf * self.balance[l] * self.obj_ratio) + return loss + + def get_near_points(self, x, y, i, j): + sub_x = x - i + sub_y = y - j + if sub_x > 0.5 and sub_y > 0.5: + return [[0, 0], [1, 0], [0, 1]] + elif sub_x < 0.5 and sub_y > 0.5: + return [[0, 0], [-1, 0], [0, 1]] + elif sub_x < 0.5 and sub_y < 0.5: + return [[0, 0], [-1, 0], [0, -1]] + else: + return [[0, 0], [1, 0], [0, -1]] + + def get_target(self, l, targets, anchors, in_h, in_w): + #-----------------------------------------------------# + # 计算一共有多少张图片 + #-----------------------------------------------------# + bs = len(targets) + #-----------------------------------------------------# + # 用于选取哪些先验框不包含物体 + # bs, 3, 20, 20 + #-----------------------------------------------------# + noobj_mask = torch.ones(bs, len(self.anchors_mask[l]), in_h, in_w, requires_grad = False) + #-----------------------------------------------------# + # 帮助找到每一个先验框最对应的真实框 + #-----------------------------------------------------# + box_best_ratio = torch.zeros(bs, len(self.anchors_mask[l]), in_h, in_w, requires_grad = False) + #-----------------------------------------------------# + # batch_size, 3, 20, 20, 5 + num_classes + #-----------------------------------------------------# + y_true = torch.zeros(bs, len(self.anchors_mask[l]), in_h, in_w, self.bbox_attrs, requires_grad = False) + for b in range(bs): + if len(targets[b])==0: + continue + batch_target = torch.zeros_like(targets[b]) + #-------------------------------------------------------# + # 计算出正样本在特征层上的中心点 + # 获得真实框相对于特征层的大小 + #-------------------------------------------------------# + batch_target[:, [0,2]] = targets[b][:, [0,2]] * in_w + batch_target[:, [1,3]] = targets[b][:, [1,3]] * in_h + batch_target[:, 4] = targets[b][:, 4] + batch_target = batch_target.cpu() + + #-----------------------------------------------------------------------------# + # batch_target : num_true_box, 5 + # batch_target[:, 2:4] : num_true_box, 2 + # torch.unsqueeze(batch_target[:, 2:4], 1) : num_true_box, 1, 2 + # anchors : 9, 2 + # torch.unsqueeze(torch.FloatTensor(anchors), 0) : 1, 9, 2 + # ratios_of_gt_anchors : num_true_box, 9, 2 + # ratios_of_anchors_gt : num_true_box, 9, 2 + # + # ratios : num_true_box, 9, 4 + # max_ratios : num_true_box, 9 + # max_ratios每一个真实框和每一个先验框的最大宽高比！ + #------------------------------------------------------------------------------# + ratios_of_gt_anchors = torch.unsqueeze(batch_target[:, 2:4], 1) / torch.unsqueeze(torch.FloatTensor(anchors), 0) + ratios_of_anchors_gt = torch.unsqueeze(torch.FloatTensor(anchors), 0) / torch.unsqueeze(batch_target[:, 2:4], 1) + ratios = torch.cat([ratios_of_gt_anchors, ratios_of_anchors_gt], dim = -1) + max_ratios, _ = torch.max(ratios, dim = -1) + + for t, ratio in enumerate(max_ratios): + #-------------------------------------------------------# + # ratio : 9 + #-------------------------------------------------------# + over_threshold = ratio < self.threshold + over_threshold[torch.argmin(ratio)] = True + for k, mask in enumerate(self.anchors_mask[l]): + if not over_threshold[mask]: + continue + #----------------------------------------# + # 获得真实框属于哪个网格点 + # x 1.25 => 1 + # y 3.75 => 3 + #----------------------------------------# + i = torch.floor(batch_target[t, 0]).long() + j = torch.floor(batch_target[t, 1]).long() + + offsets = self.get_near_points(batch_target[t, 0], batch_target[t, 1], i, j) + for offset in offsets: + local_i = i + offset[0] + local_j = j + offset[1] + + if local_i >= in_w or local_i < 0 or local_j >= in_h or local_j < 0: + continue + + if box_best_ratio[b, k, local_j, local_i] != 0: + if box_best_ratio[b, k, local_j, local_i] > ratio[mask]: + y_true[b, k, local_j, local_i, :] = 0 + else: + continue + + #----------------------------------------# + # 取出真实框的种类 + #----------------------------------------# + c = batch_target[t, 4].long() + + #----------------------------------------# + # noobj_mask代表无目标的特征点 + #----------------------------------------# + noobj_mask[b, k, local_j, local_i] = 0 + #----------------------------------------# + # tx、ty代表中心调整参数的真实值 + #----------------------------------------# + y_true[b, k, local_j, local_i, 0] = batch_target[t, 0] + y_true[b, k, local_j, local_i, 1] = batch_target[t, 1] + y_true[b, k, local_j, local_i, 2] = batch_target[t, 2] + y_true[b, k, local_j, local_i, 3] = batch_target[t, 3] + y_true[b, k, local_j, local_i, 4] = 1 + y_true[b, k, local_j, local_i, c + 5] = 1 + #----------------------------------------# + # 获得当前先验框最好的比例 + #----------------------------------------# + box_best_ratio[b, k, local_j, local_i] = ratio[mask] + + return y_true, noobj_mask + + def get_pred_boxes(self, l, x, y, h, w, targets, scaled_anchors, in_h, in_w): + #-----------------------------------------------------# + # 计算一共有多少张图片 + #-----------------------------------------------------# + bs = len(targets) + + #-----------------------------------------------------# + # 生成网格，先验框中心，网格左上角 + #-----------------------------------------------------# + grid_x = torch.linspace(0, in_w - 1, in_w).repeat(in_h, 1).repeat( + int(bs * len(self.anchors_mask[l])), 1, 1).view(x.shape).type_as(x) + grid_y = torch.linspace(0, in_h - 1, in_h).repeat(in_w, 1).t().repeat( + int(bs * len(self.anchors_mask[l])), 1, 1).view(y.shape).type_as(x) + + # 生成先验框的宽高 + scaled_anchors_l = np.array(scaled_anchors)[self.anchors_mask[l]] + anchor_w = torch.Tensor(scaled_anchors_l).index_select(1, torch.LongTensor([0])).type_as(x) + anchor_h = torch.Tensor(scaled_anchors_l).index_select(1, torch.LongTensor([1])).type_as(x) + + anchor_w = anchor_w.repeat(bs, 1).repeat(1, 1, in_h * in_w).view(w.shape) + anchor_h = anchor_h.repeat(bs, 1).repeat(1, 1, in_h * in_w).view(h.shape) + #-------------------------------------------------------# + # 计算调整后的先验框中心与宽高 + #-------------------------------------------------------# + pred_boxes_x = torch.unsqueeze(x * 2. - 0.5 + grid_x, -1) + pred_boxes_y = torch.unsqueeze(y * 2. - 0.5 + grid_y, -1) + pred_boxes_w = torch.unsqueeze((w * 2) ** 2 * anchor_w, -1) + pred_boxes_h = torch.unsqueeze((h * 2) ** 2 * anchor_h, -1) + pred_boxes = torch.cat([pred_boxes_x, pred_boxes_y, pred_boxes_w, pred_boxes_h], dim = -1) + return pred_boxes + +def is_parallel(model): + # Returns True if model is of type DP or DDP + return type(model) in (nn.parallel.DataParallel, nn.parallel.DistributedDataParallel) + +def de_parallel(model): + # De-parallelize a model: returns single-GPU model if model is of type DP or DDP + return model.module if is_parallel(model) else model + +def copy_attr(a, b, include=(), exclude=()): + # Copy attributes from b to a, options to only include [...] and to exclude [...] + for k, v in b.__dict__.items(): + if (len(include) and k not in include) or k.startswith('_') or k in exclude: + continue + else: + setattr(a, k, v) + +class ModelEMA: + """ Updated Exponential Moving Average (EMA) from https://github.com/rwightman/pytorch-image-models + Keeps a moving average of everything in the model state_dict (parameters and buffers) + For EMA details see https://www.tensorflow.org/api_docs/python/tf/train/ExponentialMovingAverage + """ + + def __init__(self, model, decay=0.9999, tau=2000, updates=0): + # Create EMA + self.ema = deepcopy(de_parallel(model)).eval() # FP32 EMA + # if next(model.parameters()).device.type != 'cpu': + # self.ema.half() # FP16 EMA + self.updates = updates # number of EMA updates + self.decay = lambda x: decay * (1 - math.exp(-x / tau)) # decay exponential ramp (to help early epochs) + for p in self.ema.parameters(): + p.requires_grad_(False) + + def update(self, model): + # Update EMA parameters + with torch.no_grad(): + self.updates += 1 + d = self.decay(self.updates) + + msd = de_parallel(model).state_dict() # model state_dict + for k, v in self.ema.state_dict().items(): + if v.dtype.is_floating_point: + v *= d + v += (1 - d) * msd[k].detach() + + def update_attr(self, model, include=(), exclude=('process_group', 'reducer')): + # Update EMA attributes + copy_attr(self.ema, model, include, exclude) + +def weights_init(net, init_type='normal', init_gain = 0.02): + def init_func(m): + classname = m.__class__.__name__ + if hasattr(m, 'weight') and classname.find('Conv') != -1: + if init_type == 'normal': + torch.nn.init.normal_(m.weight.data, 0.0, init_gain) + elif init_type == 'xavier': + torch.nn.init.xavier_normal_(m.weight.data, gain=init_gain) + elif init_type == 'kaiming': + torch.nn.init.kaiming_normal_(m.weight.data, a=0, mode='fan_in') + elif init_type == 'orthogonal': + torch.nn.init.orthogonal_(m.weight.data, gain=init_gain) + else: + raise NotImplementedError('initialization method [%s] is not implemented' % init_type) + elif classname.find('BatchNorm2d') != -1: + torch.nn.init.normal_(m.weight.data, 1.0, 0.02) + torch.nn.init.constant_(m.bias.data, 0.0) + print('initialize network with %s type' % init_type) + net.apply(init_func) + +def get_lr_scheduler(lr_decay_type, lr, min_lr, total_iters, warmup_iters_ratio = 0.05, warmup_lr_ratio = 0.1, no_aug_iter_ratio = 0.05, step_num = 10): + def yolox_warm_cos_lr(lr, min_lr, total_iters, warmup_total_iters, warmup_lr_start, no_aug_iter, iters): + if iters <= warmup_total_iters: + # lr = (lr - warmup_lr_start) * iters / float(warmup_total_iters) + warmup_lr_start + lr = (lr - warmup_lr_start) * pow(iters / float(warmup_total_iters), 2 + ) + warmup_lr_start + elif iters >= total_iters - no_aug_iter: + lr = min_lr + else: + lr = min_lr + 0.5 * (lr - min_lr) * ( + 1.0 + + math.cos( + math.pi + * (iters - warmup_total_iters) + / (total_iters - warmup_total_iters - no_aug_iter) + ) + ) + return lr + + def step_lr(lr, decay_rate, step_size, iters): + if step_size < 1: + raise ValueError("step_size must above 1.") + n = iters // step_size + out_lr = lr * decay_rate ** n + return out_lr + + if lr_decay_type == "cos": + warmup_total_iters = min(max(warmup_iters_ratio * total_iters, 1), 3) + warmup_lr_start = max(warmup_lr_ratio * lr, 1e-6) + no_aug_iter = min(max(no_aug_iter_ratio * total_iters, 1), 15) + func = partial(yolox_warm_cos_lr ,lr, min_lr, total_iters, warmup_total_iters, warmup_lr_start, no_aug_iter) + else: + decay_rate = (min_lr / lr) ** (1 / (step_num - 1)) + step_size = total_iters / step_num + func = partial(step_lr, lr, decay_rate, step_size) + + return func + +def set_optimizer_lr(optimizer, lr_scheduler_func, epoch): + lr = lr_scheduler_func(epoch) + for param_group in optimizer.param_groups: + param_group['lr'] = lr diff --git a/predict.py b/predict.py new file mode 100644 index 0000000..a7cc1c5 --- /dev/null +++ b/predict.py @@ -0,0 +1,192 @@ +#-----------------------------------------------------------------------# +# predict.py将单张图片预测、摄像头检测、FPS测试和目录遍历检测等功能 +# 整合到了一个py文件中，通过指定mode进行模式的修改。 +#-----------------------------------------------------------------------# +import time + +import cv2 +import numpy as np +from PIL import Image + +from yolo import YOLO, YOLO_ONNX + +if __name__ == "__main__": + #----------------------------------------------------------------------------------------------------------# + # mode用于指定测试的模式： + # 'predict' 表示单张图片预测，如果想对预测过程进行修改，如保存图片，截取对象等，可以先看下方详细的注释 + # 'video' 表示视频检测，可调用摄像头或者视频进行检测，详情查看下方注释。 + # 'fps' 表示测试fps，使用的图片是img里面的street.jpg，详情查看下方注释。 + # 'dir_predict' 表示遍历文件夹进行检测并保存。默认遍历img文件夹，保存img_out文件夹，详情查看下方注释。 + # 'heatmap' 表示进行预测结果的热力图可视化，详情查看下方注释。 + # 'export_onnx' 表示将模型导出为onnx，需要pytorch1.7.1以上。 + # 'predict_onnx' 表示利用导出的onnx模型进行预测，相关参数的修改在yolo.py_423行左右处的YOLO_ONNX + #----------------------------------------------------------------------------------------------------------# + mode = "predict" + #-------------------------------------------------------------------------# + # crop 指定了是否在单张图片预测后对目标进行截取 + # count 指定了是否进行目标的计数 + # crop、count仅在mode='predict'时有效 + #-------------------------------------------------------------------------# + crop = False + count = False + #----------------------------------------------------------------------------------------------------------# + # video_path 用于指定视频的路径，当video_path=0时表示检测摄像头 + # 想要检测视频，则设置如video_path = "xxx.mp4"即可，代表读取出根目录下的xxx.mp4文件。 + # video_save_path 表示视频保存的路径，当video_save_path=""时表示不保存 + # 想要保存视频，则设置如video_save_path = "yyy.mp4"即可，代表保存为根目录下的yyy.mp4文件。 + # video_fps 用于保存的视频的fps + # + # video_path、video_save_path和video_fps仅在mode='video'时有效 + # 保存视频时需要ctrl+c退出或者运行到最后一帧才会完成完整的保存步骤。 + #----------------------------------------------------------------------------------------------------------# + video_path = 0 + video_save_path = "" + video_fps = 25.0 + #----------------------------------------------------------------------------------------------------------# + # test_interval 用于指定测量fps的时候，图片检测的次数。理论上test_interval越大，fps越准确。 + # fps_image_path 用于指定测试的fps图片 + # + # test_interval和fps_image_path仅在mode='fps'有效 + #----------------------------------------------------------------------------------------------------------# + test_interval = 100 + fps_image_path = "img/street.jpg" + #-------------------------------------------------------------------------# + # dir_origin_path 指定了用于检测的图片的文件夹路径 + # dir_save_path 指定了检测完图片的保存路径 + # + # dir_origin_path和dir_save_path仅在mode='dir_predict'时有效 + #-------------------------------------------------------------------------# + dir_origin_path = "application\logs\logData\img" + dir_save_path = "application\logs\logData\save" + #-------------------------------------------------------------------------# + # heatmap_save_path 热力图的保存路径，默认保存在model_data下 + # + # heatmap_save_path仅在mode='heatmap'有效 + #-------------------------------------------------------------------------# + heatmap_save_path = "model_data/heatmap_vision.png" + #-------------------------------------------------------------------------# + # simplify 使用Simplify onnx + # onnx_save_path 指定了onnx的保存路径 + #-------------------------------------------------------------------------# + simplify = True + onnx_save_path = "model_data/models.onnx" + + if mode != "predict_onnx": + yolo = YOLO() + else: + yolo = YOLO_ONNX() + + if mode == "predict": + ''' + 1、如果想要进行检测完的图片的保存，利用r_image.save("img.jpg")即可保存，直接在predict.py里进行修改即可。 + 2、如果想要获得预测框的坐标，可以进入yolo.detect_image函数，在绘图部分读取top，left，bottom，right这四个值。 + 3、如果想要利用预测框截取下目标，可以进入yolo.detect_image函数，在绘图部分利用获取到的top，left，bottom，right这四个值 + 在原图上利用矩阵的方式进行截取。 + 4、如果想要在预测图上写额外的字，比如检测到的特定目标的数量，可以进入yolo.detect_image函数，在绘图部分对predicted_class进行判断， + 比如判断if predicted_class == 'car': 即可判断当前目标是否为车，然后记录数量即可。利用draw.text即可写字。 + ''' + while True: + img = input('Input image filename:') + try: + image = Image.open(img) + except: + print('Open Error! Try again!') + continue + else: + r_image = yolo.detect_image(image, crop = crop, count=count) + r_image.show() + + elif mode == "video": + capture = cv2.VideoCapture(video_path) + if video_save_path!="": + fourcc = cv2.VideoWriter_fourcc(*'XVID') + size = (int(capture.get(cv2.CAP_PROP_FRAME_WIDTH)), int(capture.get(cv2.CAP_PROP_FRAME_HEIGHT))) + out = cv2.VideoWriter(video_save_path, fourcc, video_fps, size) + + ref, frame = capture.read() + if not ref: + raise ValueError("未能正确读取摄像头（视频），请注意是否正确安装摄像头（是否正确填写视频路径）。") + + fps = 0.0 + while(True): + t1 = time.time() + # 读取某一帧 + ref, frame = capture.read() + if not ref: + break + # 格式转变，BGRtoRGB + frame = cv2.cvtColor(frame,cv2.COLOR_BGR2RGB) + # 转变成Image + frame = Image.fromarray(np.uint8(frame)) + # 进行检测 + frame = np.array(yolo.detect_image(frame)) + # RGBtoBGR满足opencv显示格式 + frame = cv2.cvtColor(frame,cv2.COLOR_RGB2BGR) + + fps = ( fps + (1./(time.time()-t1)) ) / 2 + print("fps= %.2f"%(fps)) + frame = cv2.putText(frame, "fps= %.2f"%(fps), (0, 40), cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 255, 0), 2) + + cv2.imshow("video",frame) + c= cv2.waitKey(1) & 0xff + if video_save_path!="": + out.write(frame) + + if c==27: + capture.release() + break + + print("Video Detection Done!") + capture.release() + if video_save_path!="": + print("Save processed video to the path :" + video_save_path) + out.release() + cv2.destroyAllWindows() + + elif mode == "fps": + img = Image.open(fps_image_path) + tact_time = yolo.get_FPS(img, test_interval) + print(str(tact_time) + ' seconds, ' + str(1/tact_time) + 'FPS, @batch_size 1') + + elif mode == "dir_predict": + import os + + from tqdm import tqdm + + img_names = os.listdir(dir_origin_path) + for img_name in tqdm(img_names): + if img_name.lower().endswith(('.bmp', '.dib', '.png', '.jpg', '.jpeg', '.pbm', '.pgm', '.ppm', '.tif', '.tiff')): + image_path = os.path.join(dir_origin_path, img_name) + image = Image.open(image_path) + r_image = yolo.detect_image(image) + if not os.path.exists(dir_save_path): + os.makedirs(dir_save_path) + r_image.save(os.path.join(dir_save_path, img_name.replace(".jpg", ".png")), quality=95, subsampling=0) + + elif mode == "heatmap": + while True: + img = input('Input image filename:') + try: + image = Image.open(img) + except: + print('Open Error! Try again!') + continue + else: + yolo.detect_heatmap(image, heatmap_save_path) + + elif mode == "export_onnx": + yolo.convert_to_onnx(simplify, onnx_save_path) + + elif mode == "predict_onnx": + while True: + img = input('Input image filename:') + try: + image = Image.open(img) + except: + print('Open Error! Try again!') + continue + else: + r_image = yolo.detect_image(image) + r_image.show() + else: + raise AssertionError("Please specify the correct mode: 'predict', 'video', 'fps', 'heatmap', 'export_onnx', 'dir_predict'.") diff --git a/summary.py b/summary.py new file mode 100644 index 0000000..67c3006 --- /dev/null +++ b/summary.py @@ -0,0 +1,32 @@ +#--------------------------------------------# +# 该部分代码用于看网络结构 +#--------------------------------------------# +import torch +from thop import clever_format, profile +from torchsummary import summary + +from nets.yolo import YoloBody + +if __name__ == "__main__": + input_shape = [640, 640] + anchors_mask = [[6, 7, 8], [3, 4, 5], [0, 1, 2]] + num_classes = 80 + backbone = 'cspdarknet' + phi = 'l' + + device = torch.device("cuda" if torch.cuda.is_available() else "cpu") + m = YoloBody(anchors_mask, num_classes, phi, backbone=backbone).to(device) + summary(m, (3, input_shape[0], input_shape[1])) + + dummy_input = torch.randn(1, 3, input_shape[0], input_shape[1]).to(device) + flops, params = profile(m.to(device), (dummy_input, ), verbose=False) + #--------------------------------------------------------# + # flops * 2是因为profile没有将卷积作为两个operations + # 有些论文将卷积算乘法、加法两个operations。此时乘2 + # 有些论文只考虑乘法的运算次数，忽略加法。此时不乘2 + # 本代码选择乘2，参考YOLOX。 + #--------------------------------------------------------# + flops = flops * 2 + flops, params = clever_format([flops, params], "%.3f") + print('Total GFLOPS: %s' % (flops)) + print('Total params: %s' % (params)) diff --git a/test.py b/test.py deleted file mode 100644 index e69de29..0000000 diff --git a/train.py b/train.py index acdafc2..6c4e889 100644 --- a/train.py +++ b/train.py @@ -181,8 +181,8 @@ if __name__ == "__main__": # Adam可以使用相对较小的UnFreeze_Epoch # Unfreeze_batch_size 模型在解冻后的batch_size #------------------------------------------------------------------# - UnFreeze_Epoch = 500 - Unfreeze_batch_size = 6 + UnFreeze_Epoch = 100 + Unfreeze_batch_size = 4 #------------------------------------------------------------------# # Freeze_Train 是否进行冻结训练 # 默认先冻结主干训练后解冻训练。 @@ -237,14 +237,14 @@ if __name__ == "__main__": # 开启后会加快数据读取速度，但是会占用更多内存 # 内存较小的电脑可以设置为2或者0 #------------------------------------------------------------------# - num_workers = 20 + num_workers = 6 #------------------------------------------------------# # train_annotation_path 训练图片路径和标签 # val_annotation_path 验证图片路径和标签 #------------------------------------------------------# - train_annotation_path = '2007_train.txt' - val_annotation_path = '2007_val.txt' + train_annotation_path = 'model_data/2007_train.txt' + val_annotation_path = 'model_data/2007_val.txt' seed_everything(seed) #------------------------------------------------------# @@ -261,6 +261,7 @@ if __name__ == "__main__": print("Gpu Device Count : ", ngpus_per_node) else: device = torch.device('cuda' if torch.cuda.is_available() else 'cpu') + print("\033[1;33;44mRuning on {}\033[0m".format(device)) local_rank = 0 rank = 0 diff --git a/utils/get_map.py b/utils/get_map.py new file mode 100644 index 0000000..ccda5eb --- /dev/null +++ b/utils/get_map.py @@ -0,0 +1,138 @@ +import os +import xml.etree.ElementTree as ET + +from PIL import Image +from tqdm import tqdm + +from utils.utils import get_classes +from utils.utils_map import get_coco_map, get_map +from yolo import YOLO + +if __name__ == "__main__": + ''' + Recall和Precision不像AP是一个面积的概念，因此在门限值（Confidence）不同时，网络的Recall和Precision值是不同的。 + 默认情况下，本代码计算的Recall和Precision代表的是当门限值（Confidence）为0.5时，所对应的Recall和Precision值。 + + 受到mAP计算原理的限制，网络在计算mAP时需要获得近乎所有的预测框，这样才可以计算不同门限条件下的Recall和Precision值 + 因此，本代码获得的map_out/detection-results/里面的txt的框的数量一般会比直接predict多一些，目的是列出所有可能的预测框， + ''' + #------------------------------------------------------------------------------------------------------------------# + # map_mode用于指定该文件运行时计算的内容 + # map_mode为0代表整个map计算流程，包括获得预测结果、获得真实框、计算VOC_map。 + # map_mode为1代表仅仅获得预测结果。 + # map_mode为2代表仅仅获得真实框。 + # map_mode为3代表仅仅计算VOC_map。 + # map_mode为4代表利用COCO工具箱计算当前数据集的0.50:0.95map。需要获得预测结果、获得真实框后并安装pycocotools才行 + #-------------------------------------------------------------------------------------------------------------------# + map_mode = 0 + #--------------------------------------------------------------------------------------# + # 此处的classes_path用于指定需要测量VOC_map的类别 + # 一般情况下与训练和预测所用的classes_path一致即可 + #--------------------------------------------------------------------------------------# + classes_path = 'model_data/voc_classes.txt' + #--------------------------------------------------------------------------------------# + # MINOVERLAP用于指定想要获得的mAP0.x，mAP0.x的意义是什么请同学们百度一下。 + # 比如计算mAP0.75，可以设定MINOVERLAP = 0.75。 + # + # 当某一预测框与真实框重合度大于MINOVERLAP时，该预测框被认为是正样本，否则为负样本。 + # 因此MINOVERLAP的值越大，预测框要预测的越准确才能被认为是正样本，此时算出来的mAP值越低， + #--------------------------------------------------------------------------------------# + MINOVERLAP = 0.5 + #--------------------------------------------------------------------------------------# + # 受到mAP计算原理的限制，网络在计算mAP时需要获得近乎所有的预测框，这样才可以计算mAP + # 因此，confidence的值应当设置的尽量小进而获得全部可能的预测框。 + # + # 该值一般不调整。因为计算mAP需要获得近乎所有的预测框，此处的confidence不能随便更改。 + # 想要获得不同门限值下的Recall和Precision值，请修改下方的score_threhold。 + #--------------------------------------------------------------------------------------# + confidence = 0.001 + #--------------------------------------------------------------------------------------# + # 预测时使用到的非极大抑制值的大小，越大表示非极大抑制越不严格。 + # + # 该值一般不调整。 + #--------------------------------------------------------------------------------------# + nms_iou = 0.5 + #---------------------------------------------------------------------------------------------------------------# + # Recall和Precision不像AP是一个面积的概念，因此在门限值不同时，网络的Recall和Precision值是不同的。 + # + # 默认情况下，本代码计算的Recall和Precision代表的是当门限值为0.5（此处定义为score_threhold）时所对应的Recall和Precision值。 + # 因为计算mAP需要获得近乎所有的预测框，上面定义的confidence不能随便更改。 + # 这里专门定义一个score_threhold用于代表门限值，进而在计算mAP时找到门限值对应的Recall和Precision值。 + #---------------------------------------------------------------------------------------------------------------# + score_threhold = 0.5 + #-------------------------------------------------------# + # map_vis用于指定是否开启VOC_map计算的可视化 + #-------------------------------------------------------# + map_vis = False + #-------------------------------------------------------# + # 指向VOC数据集所在的文件夹 + # 默认指向根目录下的VOC数据集 + #-------------------------------------------------------# + VOCdevkit_path = 'VOCdevkit' + #-------------------------------------------------------# + # 结果输出的文件夹，默认为map_out + #-------------------------------------------------------# + map_out_path = 'map_out' + + image_ids = open(os.path.join(VOCdevkit_path, "VOC2007/ImageSets/Main/test.txt")).read().strip().split() + + if not os.path.exists(map_out_path): + os.makedirs(map_out_path) + if not os.path.exists(os.path.join(map_out_path, 'ground-truth')): + os.makedirs(os.path.join(map_out_path, 'ground-truth')) + if not os.path.exists(os.path.join(map_out_path, 'detection-results')): + os.makedirs(os.path.join(map_out_path, 'detection-results')) + if not os.path.exists(os.path.join(map_out_path, 'images-optional')): + os.makedirs(os.path.join(map_out_path, 'images-optional')) + + class_names, _ = get_classes(classes_path) + + if map_mode == 0 or map_mode == 1: + print("Load model.") + yolo = YOLO(confidence = confidence, nms_iou = nms_iou) + print("Load model done.") + + print("Get predict result.") + for image_id in tqdm(image_ids): + image_path = os.path.join(VOCdevkit_path, "VOC2007/JPEGImages/"+image_id+".jpg") + image = Image.open(image_path) + if map_vis: + image.save(os.path.join(map_out_path, "images-optional/" + image_id + ".jpg")) + yolo.get_map_txt(image_id, image, class_names, map_out_path) + print("Get predict result done.") + + if map_mode == 0 or map_mode == 2: + print("Get ground truth result.") + for image_id in tqdm(image_ids): + with open(os.path.join(map_out_path, "ground-truth/"+image_id+".txt"), "w") as new_f: + root = ET.parse(os.path.join(VOCdevkit_path, "VOC2007/Annotations/"+image_id+".xml")).getroot() + for obj in root.findall('object'): + difficult_flag = False + if obj.find('difficult')!=None: + difficult = obj.find('difficult').text + if int(difficult)==1: + difficult_flag = True + obj_name = obj.find('name').text + if obj_name not in class_names: + continue + bndbox = obj.find('bndbox') + left = bndbox.find('xmin').text + top = bndbox.find('ymin').text + right = bndbox.find('xmax').text + bottom = bndbox.find('ymax').text + + if difficult_flag: + new_f.write("%s %s %s %s %s difficult\n" % (obj_name, left, top, right, bottom)) + else: + new_f.write("%s %s %s %s %s\n" % (obj_name, left, top, right, bottom)) + print("Get ground truth result done.") + + if map_mode == 0 or map_mode == 3: + print("Get map.") + get_map(MINOVERLAP, True, score_threhold = score_threhold, path = map_out_path) + print("Get map done.") + + if map_mode == 4: + print("Get map.") + get_coco_map(class_names = class_names, path = map_out_path) + print("Get map done.") diff --git a/utils/summary.py b/utils/summary.py new file mode 100644 index 0000000..67c3006 --- /dev/null +++ b/utils/summary.py @@ -0,0 +1,32 @@ +#--------------------------------------------# +# 该部分代码用于看网络结构 +#--------------------------------------------# +import torch +from thop import clever_format, profile +from torchsummary import summary + +from nets.yolo import YoloBody + +if __name__ == "__main__": + input_shape = [640, 640] + anchors_mask = [[6, 7, 8], [3, 4, 5], [0, 1, 2]] + num_classes = 80 + backbone = 'cspdarknet' + phi = 'l' + + device = torch.device("cuda" if torch.cuda.is_available() else "cpu") + m = YoloBody(anchors_mask, num_classes, phi, backbone=backbone).to(device) + summary(m, (3, input_shape[0], input_shape[1])) + + dummy_input = torch.randn(1, 3, input_shape[0], input_shape[1]).to(device) + flops, params = profile(m.to(device), (dummy_input, ), verbose=False) + #--------------------------------------------------------# + # flops * 2是因为profile没有将卷积作为两个operations + # 有些论文将卷积算乘法、加法两个operations。此时乘2 + # 有些论文只考虑乘法的运算次数，忽略加法。此时不乘2 + # 本代码选择乘2，参考YOLOX。 + #--------------------------------------------------------# + flops = flops * 2 + flops, params = clever_format([flops, params], "%.3f") + print('Total GFLOPS: %s' % (flops)) + print('Total params: %s' % (params)) diff --git a/voc_annotation.py b/utils/voc_annotation.py similarity index 93% rename from voc_annotation.py rename to utils/voc_annotation.py index 582a4fe..53d5f65 100644 --- a/voc_annotation.py +++ b/utils/voc_annotation.py @@ -3,9 +3,13 @@ import random import xml.etree.ElementTree as ET import numpy as np +import sys,os +sys.path.append(os.path.dirname(os.path.dirname(os.path.abspath(__file__)))) -from utils.utils import get_classes - +from utils import get_classes +import configparser +conf=configparser.ConfigParser() +conf.read('config.ini',encoding='utf-8') #--------------------------------------------------------------------------------------------------------------------------------# # annotation_mode用于指定该文件运行时计算的内容 # annotation_mode为0代表整个标签处理过程，包括获得VOCdevkit/VOC2007/ImageSets里面的txt以及训练用的2007_train.txt、2007_val.txt @@ -20,7 +24,7 @@ annotation_mode = 0 # 那么就是因为classes没有设定正确 # 仅在annotation_mode为0和2的时候有效 #-------------------------------------------------------------------# -classes_path = 'trainYolov5-v6/model_data/voc_classes.txt' +classes_path = conf.get('dataset', 'classes_path') #--------------------------------------------------------------------------------------------------------------------------------# # trainval_percent用于指定(训练集+验证集)与测试集的比例，默认情况下 (训练集+验证集):测试集 = 9:1 # train_percent用于指定(训练集+验证集)中训练集与验证集的比例，默认情况下 训练集:验证集 = 9:1 @@ -32,7 +36,7 @@ train_percent = 0.9 # 指向VOC数据集所在的文件夹 # 默认指向根目录下的VOC数据集 #-------------------------------------------------------# -VOCdevkit_path = 'Data/TrainData' +VOCdevkit_path = r'database/Train' VOCdevkit_sets = [('2007', 'train'), ('2007', 'val')] classes, _ = get_classes(classes_path) @@ -112,9 +116,9 @@ if __name__ == "__main__": type_index = 0 for year, image_set in VOCdevkit_sets: image_ids = open(os.path.join(VOCdevkit_path, 'ImageSets/Main/%s.txt'%(image_set)), encoding='utf-8').read().strip().split() - list_file = open('%s_%s.txt'%(year, image_set), 'w', encoding='utf-8') + list_file = open('model_data/%s_%s.txt'%(year, image_set), 'w', encoding='utf-8') for image_id in image_ids: - list_file.write('%s/JPEGImages/%s.jpg'%(os.path.abspath(VOCdevkit_path), image_id)) + list_file.write('%s/JPEGImages/%s.png'%(os.path.abspath(VOCdevkit_path), image_id)) convert_annotation(year, image_id, list_file) list_file.write('\n') diff --git a/utils/yolo.py b/utils/yolo.py new file mode 100644 index 0000000..d6e24cc --- /dev/null +++ b/utils/yolo.py @@ -0,0 +1,663 @@ +import colorsys +import os +import time + +import cv2 +import numpy as np +import torch +import torch.nn as nn +from PIL import ImageDraw, ImageFont, Image + +from nets.yolo import YoloBody +from utils.utils import (cvtColor, get_anchors, get_classes, preprocess_input, + resize_image, show_config) +from utils.utils_bbox import DecodeBox, DecodeBoxNP + +''' +训练自己的数据集必看注释！ +''' +class YOLO(object): + _defaults = { + #--------------------------------------------------------------------------# + # 使用自己训练好的模型进行预测一定要修改model_path和classes_path！ + # model_path指向logs文件夹下的权值文件，classes_path指向model_data下的txt + # + # 训练好后logs文件夹下存在多个权值文件，选择验证集损失较低的即可。 + # 验证集损失较低不代表mAP较高，仅代表该权值在验证集上泛化性能较好。 + # 如果出现shape不匹配，同时要注意训练时的model_path和classes_path参数的修改 + #--------------------------------------------------------------------------# + "model_path" : r'logs-yolov5\1.pth', + "classes_path" : 'trainYolov5-v6\\model_data/coco_classes.txt', + #---------------------------------------------------------------------# + # anchors_path代表先验框对应的txt文件，一般不修改。 + # anchors_mask用于帮助代码找到对应的先验框，一般不修改。 + #---------------------------------------------------------------------# + "anchors_path" : 'trainYolov5-v6\\model_data/yolo_anchors.txt', + "anchors_mask" : [[6, 7, 8], [3, 4, 5], [0, 1, 2]], + #---------------------------------------------------------------------# + # 输入图片的大小，必须为32的倍数。 + #---------------------------------------------------------------------# + "input_shape" : [640, 640], + #------------------------------------------------------# + # backbone cspdarknet（默认） + # convnext_tiny + # convnext_small + # swin_transfomer_tiny + #------------------------------------------------------# + "backbone" : 'cspdarknet', + #------------------------------------------------------# + # 所使用的YoloV5的版本。s、m、l、x + # 在除cspdarknet的其它主干中仅影响panet的大小 + #------------------------------------------------------# + "phi" : 's', + #---------------------------------------------------------------------# + # 只有得分大于置信度的预测框会被保留下来 + #---------------------------------------------------------------------# + "confidence" : 0.5, + #---------------------------------------------------------------------# + # 非极大抑制所用到的nms_iou大小 + #---------------------------------------------------------------------# + "nms_iou" : 0.3, + #---------------------------------------------------------------------# + # 该变量用于控制是否使用letterbox_image对输入图像进行不失真的resize， + # 在多次测试后，发现关闭letterbox_image直接resize的效果更好 + #---------------------------------------------------------------------# + "letterbox_image" : True, + #-------------------------------# + # 是否使用Cuda + # 没有GPU可以设置成False + #-------------------------------# + "cuda" : True, + } + + @classmethod + def get_defaults(cls, n): + if n in cls._defaults: + return cls._defaults[n] + else: + return "Unrecognized attribute name '" + n + "'" + + #---------------------------------------------------# + # 初始化YOLO + #---------------------------------------------------# + def __init__(self, **kwargs): + self.__dict__.update(self._defaults) + for name, value in kwargs.items(): + setattr(self, name, value) + self._defaults[name] = value + + #---------------------------------------------------# + # 获得种类和先验框的数量 + #---------------------------------------------------# + self.class_names, self.num_classes = get_classes(self.classes_path) + self.anchors, self.num_anchors = get_anchors(self.anchors_path) + self.bbox_util = DecodeBox(self.anchors, self.num_classes, (self.input_shape[0], self.input_shape[1]), self.anchors_mask) + + #---------------------------------------------------# + # 画框设置不同的颜色 + #---------------------------------------------------# + hsv_tuples = [(x / self.num_classes, 1., 1.) for x in range(self.num_classes)] + self.colors = list(map(lambda x: colorsys.hsv_to_rgb(*x), hsv_tuples)) + self.colors = list(map(lambda x: (int(x[0] * 255), int(x[1] * 255), int(x[2] * 255)), self.colors)) + self.generate() + + show_config(**self._defaults) + + #---------------------------------------------------# + # 生成模型 + #---------------------------------------------------# + def generate(self, onnx=False): + #---------------------------------------------------# + # 建立yolo模型，载入yolo模型的权重 + #---------------------------------------------------# + self.net = YoloBody(self.anchors_mask, self.num_classes, self.phi, backbone = self.backbone, input_shape = self.input_shape) + device = torch.device('cuda' if torch.cuda.is_available() else 'cpu') + self.net.load_state_dict(torch.load(self.model_path, map_location=device),strict=False) + self.net = self.net.eval() + print('{} model, and classes loaded.'.format(self.model_path)) + if not onnx: + if self.cuda: + self.net = nn.DataParallel(self.net) + self.net = self.net.cuda() + + #---------------------------------------------------# + # 检测图片 + #---------------------------------------------------# + def detect_image(self, image, crop = False, count = False): + #---------------------------------------------------# + # 计算输入图片的高和宽 + #---------------------------------------------------# + image_shape = np.array(np.shape(image)[0:2]) + #---------------------------------------------------------# + # 在这里将图像转换成RGB图像，防止灰度图在预测时报错。 + # 代码仅仅支持RGB图像的预测，所有其它类型的图像都会转化成RGB + #---------------------------------------------------------# + image = cvtColor(image) + #---------------------------------------------------------# + # 给图像增加灰条，实现不失真的resize + # 也可以直接resize进行识别 + #---------------------------------------------------------# + image_data = resize_image(image, (self.input_shape[1], self.input_shape[0]), self.letterbox_image) + #---------------------------------------------------------# + # 添加上batch_size维度 + #---------------------------------------------------------# + image_data = np.expand_dims(np.transpose(preprocess_input(np.array(image_data, dtype='float32')), (2, 0, 1)), 0) + + with torch.no_grad(): + images = torch.from_numpy(image_data) + if self.cuda: + images = images.cuda() + #---------------------------------------------------------# + # 将图像输入网络当中进行预测！ + #---------------------------------------------------------# + outputs = self.net(images) + outputs = self.bbox_util.decode_box(outputs) + #---------------------------------------------------------# + # 将预测框进行堆叠，然后进行非极大抑制 + #---------------------------------------------------------# + results = self.bbox_util.non_max_suppression(torch.cat(outputs, 1), self.num_classes, self.input_shape, + image_shape, self.letterbox_image, conf_thres = self.confidence, nms_thres = self.nms_iou) + + if results[0] is None: + return image + + top_label = np.array(results[0][:, 6], dtype = 'int32') + top_conf = results[0][:, 4] * results[0][:, 5] + top_boxes = results[0][:, :4] + #---------------------------------------------------------# + # 设置字体与边框厚度 + #---------------------------------------------------------# + font = ImageFont.truetype(font='model_data/simhei.ttf', size=np.floor(3e-2 * image.size[1] + 0.5).astype('int32')) + thickness = int(max((image.size[0] + image.size[1]) // np.mean(self.input_shape), 1)) + #---------------------------------------------------------# + # 计数 + #---------------------------------------------------------# + if count: + print("top_label:", top_label) + classes_nums = np.zeros([self.num_classes]) + for i in range(self.num_classes): + num = np.sum(top_label == i) + if num > 0: + print(self.class_names[i], " : ", num) + classes_nums[i] = num + print("classes_nums:", classes_nums) + #---------------------------------------------------------# + # 是否进行目标的裁剪 + #---------------------------------------------------------# + if crop: + for i, c in list(enumerate(top_boxes)): + top, left, bottom, right = top_boxes[i] + top = max(0, np.floor(top).astype('int32')) + left = max(0, np.floor(left).astype('int32')) + bottom = min(image.size[1], np.floor(bottom).astype('int32')) + right = min(image.size[0], np.floor(right).astype('int32')) + + dir_save_path = "img_crop" + if not os.path.exists(dir_save_path): + os.makedirs(dir_save_path) + crop_image = image.crop([left, top, right, bottom]) + crop_image.save(os.path.join(dir_save_path, "crop_" + str(i) + ".png"), quality=95, subsampling=0) + print("save crop_" + str(i) + ".png to " + dir_save_path) + #---------------------------------------------------------# + # 图像绘制 + #---------------------------------------------------------# + for i, c in list(enumerate(top_label)): + predicted_class = self.class_names[int(c)] + box = top_boxes[i] + score = top_conf[i] + + top, left, bottom, right = box + + top = max(0, np.floor(top).astype('int32')) + left = max(0, np.floor(left).astype('int32')) + bottom = min(image.size[1], np.floor(bottom).astype('int32')) + right = min(image.size[0], np.floor(right).astype('int32')) + + label = '{} {:.2f}'.format(predicted_class, score) + draw = ImageDraw.Draw(image) + label_size = draw.textsize(label, font) + label = label.encode('utf-8') + print(label, top, left, bottom, right) + + if top - label_size[1] >= 0: + text_origin = np.array([left, top - label_size[1]]) + else: + text_origin = np.array([left, top + 1]) + + for i in range(thickness): + draw.rectangle([left + i, top + i, right - i, bottom - i], outline=self.colors[c]) + draw.rectangle([tuple(text_origin), tuple(text_origin + label_size)], fill=self.colors[c]) + draw.text(text_origin, str(label,'UTF-8'), fill=(0, 0, 0), font=font) + del draw + + return image + + def get_FPS(self, image, test_interval): + image_shape = np.array(np.shape(image)[0:2]) + #---------------------------------------------------------# + # 在这里将图像转换成RGB图像，防止灰度图在预测时报错。 + # 代码仅仅支持RGB图像的预测，所有其它类型的图像都会转化成RGB + #---------------------------------------------------------# + image = cvtColor(image) + #---------------------------------------------------------# + # 给图像增加灰条，实现不失真的resize + # 也可以直接resize进行识别 + #---------------------------------------------------------# + image_data = resize_image(image, (self.input_shape[1], self.input_shape[0]), self.letterbox_image) + #---------------------------------------------------------# + # 添加上batch_size维度 + #---------------------------------------------------------# + image_data = np.expand_dims(np.transpose(preprocess_input(np.array(image_data, dtype='float32')), (2, 0, 1)), 0) + + with torch.no_grad(): + images = torch.from_numpy(image_data) + if self.cuda: + images = images.cuda() + #---------------------------------------------------------# + # 将图像输入网络当中进行预测！ + #---------------------------------------------------------# + outputs = self.net(images) + outputs = self.bbox_util.decode_box(outputs) + #---------------------------------------------------------# + # 将预测框进行堆叠，然后进行非极大抑制 + #---------------------------------------------------------# + results = self.bbox_util.non_max_suppression(torch.cat(outputs, 1), self.num_classes, self.input_shape, + image_shape, self.letterbox_image, conf_thres=self.confidence, nms_thres=self.nms_iou) + + t1 = time.time() + for _ in range(test_interval): + with torch.no_grad(): + #---------------------------------------------------------# + # 将图像输入网络当中进行预测！ + #---------------------------------------------------------# + outputs = self.net(images) + outputs = self.bbox_util.decode_box(outputs) + #---------------------------------------------------------# + # 将预测框进行堆叠，然后进行非极大抑制 + #---------------------------------------------------------# + results = self.bbox_util.non_max_suppression(torch.cat(outputs, 1), self.num_classes, self.input_shape, + image_shape, self.letterbox_image, conf_thres=self.confidence, nms_thres=self.nms_iou) + + t2 = time.time() + tact_time = (t2 - t1) / test_interval + return tact_time + + def detect_heatmap(self, image, heatmap_save_path): + import cv2 + import matplotlib.pyplot as plt + def sigmoid(x): + y = 1.0 / (1.0 + np.exp(-x)) + return y + #---------------------------------------------------------# + # 在这里将图像转换成RGB图像，防止灰度图在预测时报错。 + # 代码仅仅支持RGB图像的预测，所有其它类型的图像都会转化成RGB + #---------------------------------------------------------# + image = cvtColor(image) + #---------------------------------------------------------# + # 给图像增加灰条，实现不失真的resize + # 也可以直接resize进行识别 + #---------------------------------------------------------# + image_data = resize_image(image, (self.input_shape[1],self.input_shape[0]), self.letterbox_image) + #---------------------------------------------------------# + # 添加上batch_size维度 + #---------------------------------------------------------# + image_data = np.expand_dims(np.transpose(preprocess_input(np.array(image_data, dtype='float32')), (2, 0, 1)), 0) + + with torch.no_grad(): + images = torch.from_numpy(image_data) + if self.cuda: + images = images.cuda() + #---------------------------------------------------------# + # 将图像输入网络当中进行预测！ + #---------------------------------------------------------# + outputs = self.net(images) + + plt.imshow(image, alpha=1) + plt.axis('off') + mask = np.zeros((image.size[1], image.size[0])) + for sub_output in outputs: + sub_output = sub_output.cpu().numpy() + b, c, h, w = np.shape(sub_output) + sub_output = np.transpose(np.reshape(sub_output, [b, 3, -1, h, w]), [0, 3, 4, 1, 2])[0] + score = np.max(sigmoid(sub_output[..., 4]), -1) + score = cv2.resize(score, (image.size[0], image.size[1])) + normed_score = (score * 255).astype('uint8') + mask = np.maximum(mask, normed_score) + + plt.imshow(mask, alpha=0.5, interpolation='nearest', cmap="jet") + + plt.axis('off') + plt.subplots_adjust(top=1, bottom=0, right=1, left=0, hspace=0, wspace=0) + plt.margins(0, 0) + plt.savefig(heatmap_save_path, dpi=200, bbox_inches='tight', pad_inches = -0.1) + print("Save to the " + heatmap_save_path) + plt.show() + + def convert_to_onnx(self, simplify, model_path): + import onnx + self.generate(onnx=True) + + im = torch.zeros(1, 3, *self.input_shape).to('cpu') # image size(1, 3, 512, 512) BCHW + input_layer_names = ["images"] + output_layer_names = ["output"] + + # Export the model + print(f'Starting export with onnx {onnx.__version__}.') + torch.onnx.export(self.net, + im, + f = model_path, + verbose = False, + opset_version = 12, + training = torch.onnx.TrainingMode.EVAL, + do_constant_folding = True, + input_names = input_layer_names, + output_names = output_layer_names, + dynamic_axes = None) + + # Checks + model_onnx = onnx.load(model_path) # load onnx model + onnx.checker.check_model(model_onnx) # check onnx model + + # Simplify onnx + if simplify: + import onnxsim + print(f'Simplifying with onnx-simplifier {onnxsim.__version__}.') + model_onnx, check = onnxsim.simplify( + model_onnx, + dynamic_input_shape=False, + input_shapes=None) + assert check, 'assert check failed' + onnx.save(model_onnx, model_path) + + print('Onnx model save as {}'.format(model_path)) + + def get_map_txt(self, image_id, image, class_names, map_out_path): + f = open(os.path.join(map_out_path, "detection-results/"+image_id+".txt"), "w", encoding='utf-8') + image_shape = np.array(np.shape(image)[0:2]) + #---------------------------------------------------------# + # 在这里将图像转换成RGB图像，防止灰度图在预测时报错。 + # 代码仅仅支持RGB图像的预测，所有其它类型的图像都会转化成RGB + #---------------------------------------------------------# + image = cvtColor(image) + #---------------------------------------------------------# + # 给图像增加灰条，实现不失真的resize + # 也可以直接resize进行识别 + #---------------------------------------------------------# + image_data = resize_image(image, (self.input_shape[1], self.input_shape[0]), self.letterbox_image) + #---------------------------------------------------------# + # 添加上batch_size维度 + #---------------------------------------------------------# + image_data = np.expand_dims(np.transpose(preprocess_input(np.array(image_data, dtype='float32')), (2, 0, 1)), 0) + + with torch.no_grad(): + images = torch.from_numpy(image_data) + if self.cuda: + images = images.cuda() + #---------------------------------------------------------# + # 将图像输入网络当中进行预测！ + #---------------------------------------------------------# + outputs = self.net(images) + outputs = self.bbox_util.decode_box(outputs) + #---------------------------------------------------------# + # 将预测框进行堆叠，然后进行非极大抑制 + #---------------------------------------------------------# + results = self.bbox_util.non_max_suppression(torch.cat(outputs, 1), self.num_classes, self.input_shape, + image_shape, self.letterbox_image, conf_thres = self.confidence, nms_thres = self.nms_iou) + + if results[0] is None: + return + + top_label = np.array(results[0][:, 6], dtype = 'int32') + top_conf = results[0][:, 4] * results[0][:, 5] + top_boxes = results[0][:, :4] + + for i, c in list(enumerate(top_label)): + predicted_class = self.class_names[int(c)] + box = top_boxes[i] + score = str(top_conf[i]) + + top, left, bottom, right = box + if predicted_class not in class_names: + continue + + f.write("%s %s %s %s %s %s\n" % (predicted_class, score[:6], str(int(left)), str(int(top)), str(int(right)),str(int(bottom)))) + + f.close() + return + +class YOLO_ONNX(object): + _defaults = { + #--------------------------------------------------------------------------# + # 使用自己训练好的模型进行预测一定要修改onnx_path和classes_path！ + # onnx_path指向logs文件夹下的权值文件，classes_path指向model_data下的txt + # + # 训练好后logs文件夹下存在多个权值文件，选择验证集损失较低的即可。 + # 验证集损失较低不代表mAP较高，仅代表该权值在验证集上泛化性能较好。 + # 如果出现shape不匹配，同时要注意训练时的onnx_path和classes_path参数的修改 + #--------------------------------------------------------------------------# + "onnx_path" : 'model_data/models.onnx', + "classes_path" : 'model_data/coco_classes.txt', + #---------------------------------------------------------------------# + # anchors_path代表先验框对应的txt文件，一般不修改。 + # anchors_mask用于帮助代码找到对应的先验框，一般不修改。 + #---------------------------------------------------------------------# + "anchors_path" : 'model_data/yolo_anchors.txt', + "anchors_mask" : [[6, 7, 8], [3, 4, 5], [0, 1, 2]], + #---------------------------------------------------------------------# + # 输入图片的大小，必须为32的倍数。 + #---------------------------------------------------------------------# + "input_shape" : [640, 640], + #---------------------------------------------------------------------# + # 只有得分大于置信度的预测框会被保留下来 + #---------------------------------------------------------------------# + "confidence" : 0.5, + #---------------------------------------------------------------------# + # 非极大抑制所用到的nms_iou大小 + #---------------------------------------------------------------------# + "nms_iou" : 0.3, + #---------------------------------------------------------------------# + # 该变量用于控制是否使用letterbox_image对输入图像进行不失真的resize， + # 在多次测试后，发现关闭letterbox_image直接resize的效果更好 + #---------------------------------------------------------------------# + "letterbox_image" : True + } + + @classmethod + def get_defaults(cls, n): + if n in cls._defaults: + return cls._defaults[n] + else: + return "Unrecognized attribute name '" + n + "'" + + #---------------------------------------------------# + # 初始化YOLO + #---------------------------------------------------# + def __init__(self, **kwargs): + self.__dict__.update(self._defaults) + for name, value in kwargs.items(): + setattr(self, name, value) + self._defaults[name] = value + + import onnxruntime + self.onnx_session = onnxruntime.InferenceSession(self.onnx_path) + # 获得所有的输入node + self.input_name = self.get_input_name() + # 获得所有的输出node + self.output_name = self.get_output_name() + + #---------------------------------------------------# + # 获得种类和先验框的数量 + #---------------------------------------------------# + self.class_names, self.num_classes = self.get_classes(self.classes_path) + self.anchors, self.num_anchors = self.get_anchors(self.anchors_path) + self.bbox_util = DecodeBoxNP(self.anchors, self.num_classes, (self.input_shape[0], self.input_shape[1]), self.anchors_mask) + + #---------------------------------------------------# + # 画框设置不同的颜色 + #---------------------------------------------------# + hsv_tuples = [(x / self.num_classes, 1., 1.) for x in range(self.num_classes)] + self.colors = list(map(lambda x: colorsys.hsv_to_rgb(*x), hsv_tuples)) + self.colors = list(map(lambda x: (int(x[0] * 255), int(x[1] * 255), int(x[2] * 255)), self.colors)) + + show_config(**self._defaults) + + def get_classes(self, classes_path): + with open(classes_path, encoding='utf-8') as f: + class_names = f.readlines() + class_names = [c.strip() for c in class_names] + return class_names, len(class_names) + + def get_anchors(self, anchors_path): + '''loads the anchors from a file''' + with open(anchors_path, encoding='utf-8') as f: + anchors = f.readline() + anchors = [float(x) for x in anchors.split(',')] + anchors = np.array(anchors).reshape(-1, 2) + return anchors, len(anchors) + + def get_input_name(self): + # 获得所有的输入node + input_name=[] + for node in self.onnx_session.get_inputs(): + input_name.append(node.name) + return input_name + + def get_output_name(self): + # 获得所有的输出node + output_name=[] + for node in self.onnx_session.get_outputs(): + output_name.append(node.name) + return output_name + + def get_input_feed(self,image_tensor): + # 利用input_name获得输入的tensor + input_feed={} + for name in self.input_name: + input_feed[name]=image_tensor + return input_feed + + #---------------------------------------------------# + # 对输入图像进行resize + #---------------------------------------------------# + def resize_image(self, image, size, letterbox_image, mode='PIL'): + if mode == 'PIL': + iw, ih = image.size + w, h = size + + if letterbox_image: + scale = min(w/iw, h/ih) + nw = int(iw*scale) + nh = int(ih*scale) + + image = image.resize((nw,nh), Image.BICUBIC) + new_image = Image.new('RGB', size, (128,128,128)) + new_image.paste(image, ((w-nw)//2, (h-nh)//2)) + else: + new_image = image.resize((w, h), Image.BICUBIC) + else: + image = np.array(image) + if letterbox_image: + # 获得现在的shape + shape = np.shape(image)[:2] + # 获得输出的shape + if isinstance(size, int): + size = (size, size) + + # 计算缩放的比例 + r = min(size[0] / shape[0], size[1] / shape[1]) + + # 计算缩放后图片的高宽 + new_unpad = int(round(shape[1] * r)), int(round(shape[0] * r)) + dw, dh = size[1] - new_unpad[0], size[0] - new_unpad[1] + + # 除以2以padding到两边 + dw /= 2 + dh /= 2 + + # 对图像进行resize + if shape[::-1] != new_unpad: # resize + image = cv2.resize(image, new_unpad, interpolation=cv2.INTER_LINEAR) + top, bottom = int(round(dh - 0.1)), int(round(dh + 0.1)) + left, right = int(round(dw - 0.1)), int(round(dw + 0.1)) + + new_image = cv2.copyMakeBorder(image, top, bottom, left, right, cv2.BORDER_CONSTANT, value=(128, 128, 128)) # add border + else: + new_image = cv2.resize(image, (w, h)) + + return new_image + + def detect_image(self, image): + image_shape = np.array(np.shape(image)[0:2]) + #---------------------------------------------------------# + # 在这里将图像转换成RGB图像，防止灰度图在预测时报错。 + # 代码仅仅支持RGB图像的预测，所有其它类型的图像都会转化成RGB + #---------------------------------------------------------# + image = cvtColor(image) + + image_data = self.resize_image(image, self.input_shape, True) + #---------------------------------------------------------# + # 添加上batch_size维度 + # h, w, 3 => 3, h, w => 1, 3, h, w + #---------------------------------------------------------# + image_data = np.expand_dims(np.transpose(preprocess_input(np.array(image_data, dtype='float32')), (2, 0, 1)), 0) + + input_feed = self.get_input_feed(image_data) + outputs = self.onnx_session.run(output_names=self.output_name, input_feed=input_feed) + + feature_map_shape = [[int(j / (2 ** (i + 3))) for j in self.input_shape] for i in range(len(self.anchors_mask))][::-1] + for i in range(len(self.anchors_mask)): + outputs[i] = np.reshape(outputs[i], (1, len(self.anchors_mask[i]) * (5 + self.num_classes), feature_map_shape[i][0], feature_map_shape[i][1])) + + outputs = self.bbox_util.decode_box(outputs) + #---------------------------------------------------------# + # 将预测框进行堆叠，然后进行非极大抑制 + #---------------------------------------------------------# + results = self.bbox_util.non_max_suppression(np.concatenate(outputs, 1), self.num_classes, self.input_shape, + image_shape, self.letterbox_image, conf_thres = self.confidence, nms_thres = self.nms_iou) + + if results[0] is None: + return image + + top_label = np.array(results[0][:, 6], dtype = 'int32') + top_conf = results[0][:, 4] * results[0][:, 5] + top_boxes = results[0][:, :4] + + #---------------------------------------------------------# + # 设置字体与边框厚度 + #---------------------------------------------------------# + font = ImageFont.truetype(font='model_data/simhei.ttf', size=np.floor(3e-2 * image.size[1] + 0.5).astype('int32')) + thickness = int(max((image.size[0] + image.size[1]) // np.mean(self.input_shape), 1)) + + #---------------------------------------------------------# + # 图像绘制 + #---------------------------------------------------------# + for i, c in list(enumerate(top_label)): + predicted_class = self.class_names[int(c)] + box = top_boxes[i] + score = top_conf[i] + + top, left, bottom, right = box + + top = max(0, np.floor(top).astype('int32')) + left = max(0, np.floor(left).astype('int32')) + bottom = min(image.size[1], np.floor(bottom).astype('int32')) + right = min(image.size[0], np.floor(right).astype('int32')) + + label = '{} {:.2f}'.format(predicted_class, score) + draw = ImageDraw.Draw(image) + label_size = draw.textsize(label, font) + label = label.encode('utf-8') + print(label, top, left, bottom, right) + + if top - label_size[1] >= 0: + text_origin = np.array([left, top - label_size[1]]) + else: + text_origin = np.array([left, top + 1]) + + for i in range(thickness): + draw.rectangle([left + i, top + i, right - i, bottom - i], outline=self.colors[c]) + draw.rectangle([tuple(text_origin), tuple(text_origin + label_size)], fill=self.colors[c]) + draw.text(text_origin, str(label,'UTF-8'), fill=(0, 0, 0), font=font) + del draw + + return image \ No newline at end of file diff --git a/yolo.py b/yolo.py new file mode 100644 index 0000000..4904b6c --- /dev/null +++ b/yolo.py @@ -0,0 +1,663 @@ +import colorsys +import os +import time + +import cv2 +import numpy as np +import torch +import torch.nn as nn +from PIL import ImageDraw, ImageFont, Image + +from nets.yolo import YoloBody +from utils.utils import (cvtColor, get_anchors, get_classes, preprocess_input, + resize_image, show_config) +from utils.utils_bbox import DecodeBox, DecodeBoxNP + +''' +训练自己的数据集必看注释！ +''' +class YOLO(object): + _defaults = { + #--------------------------------------------------------------------------# + # 使用自己训练好的模型进行预测一定要修改model_path和classes_path！ + # model_path指向logs文件夹下的权值文件，classes_path指向model_data下的txt + # + # 训练好后logs文件夹下存在多个权值文件，选择验证集损失较低的即可。 + # 验证集损失较低不代表mAP较高，仅代表该权值在验证集上泛化性能较好。 + # 如果出现shape不匹配，同时要注意训练时的model_path和classes_path参数的修改 + #--------------------------------------------------------------------------# + "model_path" : r'logs-yolov5\1.pth', + "classes_path" : 'model_data/coco_classes.txt', + #---------------------------------------------------------------------# + # anchors_path代表先验框对应的txt文件，一般不修改。 + # anchors_mask用于帮助代码找到对应的先验框，一般不修改。 + #---------------------------------------------------------------------# + "anchors_path" : 'model_data/yolo_anchors.txt', + "anchors_mask" : [[6, 7, 8], [3, 4, 5], [0, 1, 2]], + #---------------------------------------------------------------------# + # 输入图片的大小，必须为32的倍数。 + #---------------------------------------------------------------------# + "input_shape" : [640, 640], + #------------------------------------------------------# + # backbone cspdarknet（默认） + # convnext_tiny + # convnext_small + # swin_transfomer_tiny + #------------------------------------------------------# + "backbone" : 'cspdarknet', + #------------------------------------------------------# + # 所使用的YoloV5的版本。s、m、l、x + # 在除cspdarknet的其它主干中仅影响panet的大小 + #------------------------------------------------------# + "phi" : 's', + #---------------------------------------------------------------------# + # 只有得分大于置信度的预测框会被保留下来 + #---------------------------------------------------------------------# + "confidence" : 0.5, + #---------------------------------------------------------------------# + # 非极大抑制所用到的nms_iou大小 + #---------------------------------------------------------------------# + "nms_iou" : 0.3, + #---------------------------------------------------------------------# + # 该变量用于控制是否使用letterbox_image对输入图像进行不失真的resize， + # 在多次测试后，发现关闭letterbox_image直接resize的效果更好 + #---------------------------------------------------------------------# + "letterbox_image" : True, + #-------------------------------# + # 是否使用Cuda + # 没有GPU可以设置成False + #-------------------------------# + "cuda" : True, + } + + @classmethod + def get_defaults(cls, n): + if n in cls._defaults: + return cls._defaults[n] + else: + return "Unrecognized attribute name '" + n + "'" + + #---------------------------------------------------# + # 初始化YOLO + #---------------------------------------------------# + def __init__(self, **kwargs): + self.__dict__.update(self._defaults) + for name, value in kwargs.items(): + setattr(self, name, value) + self._defaults[name] = value + + #---------------------------------------------------# + # 获得种类和先验框的数量 + #---------------------------------------------------# + self.class_names, self.num_classes = get_classes(self.classes_path) + self.anchors, self.num_anchors = get_anchors(self.anchors_path) + self.bbox_util = DecodeBox(self.anchors, self.num_classes, (self.input_shape[0], self.input_shape[1]), self.anchors_mask) + + #---------------------------------------------------# + # 画框设置不同的颜色 + #---------------------------------------------------# + hsv_tuples = [(x / self.num_classes, 1., 1.) for x in range(self.num_classes)] + self.colors = list(map(lambda x: colorsys.hsv_to_rgb(*x), hsv_tuples)) + self.colors = list(map(lambda x: (int(x[0] * 255), int(x[1] * 255), int(x[2] * 255)), self.colors)) + self.generate() + + show_config(**self._defaults) + + #---------------------------------------------------# + # 生成模型 + #---------------------------------------------------# + def generate(self, onnx=False): + #---------------------------------------------------# + # 建立yolo模型，载入yolo模型的权重 + #---------------------------------------------------# + self.net = YoloBody(self.anchors_mask, self.num_classes, self.phi, backbone = self.backbone, input_shape = self.input_shape) + device = torch.device('cuda' if torch.cuda.is_available() else 'cpu') + self.net.load_state_dict(torch.load(self.model_path, map_location=device),strict=False) + self.net = self.net.eval() + print('{} model, and classes loaded.'.format(self.model_path)) + if not onnx: + if self.cuda: + self.net = nn.DataParallel(self.net) + self.net = self.net.cuda() + + #---------------------------------------------------# + # 检测图片 + #---------------------------------------------------# + def detect_image(self, image, crop = False, count = False): + #---------------------------------------------------# + # 计算输入图片的高和宽 + #---------------------------------------------------# + image_shape = np.array(np.shape(image)[0:2]) + #---------------------------------------------------------# + # 在这里将图像转换成RGB图像，防止灰度图在预测时报错。 + # 代码仅仅支持RGB图像的预测，所有其它类型的图像都会转化成RGB + #---------------------------------------------------------# + image = cvtColor(image) + #---------------------------------------------------------# + # 给图像增加灰条，实现不失真的resize + # 也可以直接resize进行识别 + #---------------------------------------------------------# + image_data = resize_image(image, (self.input_shape[1], self.input_shape[0]), self.letterbox_image) + #---------------------------------------------------------# + # 添加上batch_size维度 + #---------------------------------------------------------# + image_data = np.expand_dims(np.transpose(preprocess_input(np.array(image_data, dtype='float32')), (2, 0, 1)), 0) + + with torch.no_grad(): + images = torch.from_numpy(image_data) + if self.cuda: + images = images.cuda() + #---------------------------------------------------------# + # 将图像输入网络当中进行预测！ + #---------------------------------------------------------# + outputs = self.net(images) + outputs = self.bbox_util.decode_box(outputs) + #---------------------------------------------------------# + # 将预测框进行堆叠，然后进行非极大抑制 + #---------------------------------------------------------# + results = self.bbox_util.non_max_suppression(torch.cat(outputs, 1), self.num_classes, self.input_shape, + image_shape, self.letterbox_image, conf_thres = self.confidence, nms_thres = self.nms_iou) + + if results[0] is None: + return image + + top_label = np.array(results[0][:, 6], dtype = 'int32') + top_conf = results[0][:, 4] * results[0][:, 5] + top_boxes = results[0][:, :4] + #---------------------------------------------------------# + # 设置字体与边框厚度 + #---------------------------------------------------------# + font = ImageFont.truetype(font='model_data/simhei.ttf', size=np.floor(3e-2 * image.size[1] + 0.5).astype('int32')) + thickness = int(max((image.size[0] + image.size[1]) // np.mean(self.input_shape), 1)) + #---------------------------------------------------------# + # 计数 + #---------------------------------------------------------# + if count: + print("top_label:", top_label) + classes_nums = np.zeros([self.num_classes]) + for i in range(self.num_classes): + num = np.sum(top_label == i) + if num > 0: + print(self.class_names[i], " : ", num) + classes_nums[i] = num + print("classes_nums:", classes_nums) + #---------------------------------------------------------# + # 是否进行目标的裁剪 + #---------------------------------------------------------# + if crop: + for i, c in list(enumerate(top_boxes)): + top, left, bottom, right = top_boxes[i] + top = max(0, np.floor(top).astype('int32')) + left = max(0, np.floor(left).astype('int32')) + bottom = min(image.size[1], np.floor(bottom).astype('int32')) + right = min(image.size[0], np.floor(right).astype('int32')) + + dir_save_path = "img_crop" + if not os.path.exists(dir_save_path): + os.makedirs(dir_save_path) + crop_image = image.crop([left, top, right, bottom]) + crop_image.save(os.path.join(dir_save_path, "crop_" + str(i) + ".png"), quality=95, subsampling=0) + print("save crop_" + str(i) + ".png to " + dir_save_path) + #---------------------------------------------------------# + # 图像绘制 + #---------------------------------------------------------# + for i, c in list(enumerate(top_label)): + predicted_class = self.class_names[int(c)] + box = top_boxes[i] + score = top_conf[i] + + top, left, bottom, right = box + + top = max(0, np.floor(top).astype('int32')) + left = max(0, np.floor(left).astype('int32')) + bottom = min(image.size[1], np.floor(bottom).astype('int32')) + right = min(image.size[0], np.floor(right).astype('int32')) + + label = '{} {:.2f}'.format(predicted_class, score) + draw = ImageDraw.Draw(image) + label_size = draw.textsize(label, font) + label = label.encode('utf-8') + print(label, top, left, bottom, right) + + if top - label_size[1] >= 0: + text_origin = np.array([left, top - label_size[1]]) + else: + text_origin = np.array([left, top + 1]) + + for i in range(thickness): + draw.rectangle([left + i, top + i, right - i, bottom - i], outline=self.colors[c]) + draw.rectangle([tuple(text_origin), tuple(text_origin + label_size)], fill=self.colors[c]) + draw.text(text_origin, str(label,'UTF-8'), fill=(0, 0, 0), font=font) + del draw + + return image + + def get_FPS(self, image, test_interval): + image_shape = np.array(np.shape(image)[0:2]) + #---------------------------------------------------------# + # 在这里将图像转换成RGB图像，防止灰度图在预测时报错。 + # 代码仅仅支持RGB图像的预测，所有其它类型的图像都会转化成RGB + #---------------------------------------------------------# + image = cvtColor(image) + #---------------------------------------------------------# + # 给图像增加灰条，实现不失真的resize + # 也可以直接resize进行识别 + #---------------------------------------------------------# + image_data = resize_image(image, (self.input_shape[1], self.input_shape[0]), self.letterbox_image) + #---------------------------------------------------------# + # 添加上batch_size维度 + #---------------------------------------------------------# + image_data = np.expand_dims(np.transpose(preprocess_input(np.array(image_data, dtype='float32')), (2, 0, 1)), 0) + + with torch.no_grad(): + images = torch.from_numpy(image_data) + if self.cuda: + images = images.cuda() + #---------------------------------------------------------# + # 将图像输入网络当中进行预测！ + #---------------------------------------------------------# + outputs = self.net(images) + outputs = self.bbox_util.decode_box(outputs) + #---------------------------------------------------------# + # 将预测框进行堆叠，然后进行非极大抑制 + #---------------------------------------------------------# + results = self.bbox_util.non_max_suppression(torch.cat(outputs, 1), self.num_classes, self.input_shape, + image_shape, self.letterbox_image, conf_thres=self.confidence, nms_thres=self.nms_iou) + + t1 = time.time() + for _ in range(test_interval): + with torch.no_grad(): + #---------------------------------------------------------# + # 将图像输入网络当中进行预测！ + #---------------------------------------------------------# + outputs = self.net(images) + outputs = self.bbox_util.decode_box(outputs) + #---------------------------------------------------------# + # 将预测框进行堆叠，然后进行非极大抑制 + #---------------------------------------------------------# + results = self.bbox_util.non_max_suppression(torch.cat(outputs, 1), self.num_classes, self.input_shape, + image_shape, self.letterbox_image, conf_thres=self.confidence, nms_thres=self.nms_iou) + + t2 = time.time() + tact_time = (t2 - t1) / test_interval + return tact_time + + def detect_heatmap(self, image, heatmap_save_path): + import cv2 + import matplotlib.pyplot as plt + def sigmoid(x): + y = 1.0 / (1.0 + np.exp(-x)) + return y + #---------------------------------------------------------# + # 在这里将图像转换成RGB图像，防止灰度图在预测时报错。 + # 代码仅仅支持RGB图像的预测，所有其它类型的图像都会转化成RGB + #---------------------------------------------------------# + image = cvtColor(image) + #---------------------------------------------------------# + # 给图像增加灰条，实现不失真的resize + # 也可以直接resize进行识别 + #---------------------------------------------------------# + image_data = resize_image(image, (self.input_shape[1],self.input_shape[0]), self.letterbox_image) + #---------------------------------------------------------# + # 添加上batch_size维度 + #---------------------------------------------------------# + image_data = np.expand_dims(np.transpose(preprocess_input(np.array(image_data, dtype='float32')), (2, 0, 1)), 0) + + with torch.no_grad(): + images = torch.from_numpy(image_data) + if self.cuda: + images = images.cuda() + #---------------------------------------------------------# + # 将图像输入网络当中进行预测！ + #---------------------------------------------------------# + outputs = self.net(images) + + plt.imshow(image, alpha=1) + plt.axis('off') + mask = np.zeros((image.size[1], image.size[0])) + for sub_output in outputs: + sub_output = sub_output.cpu().numpy() + b, c, h, w = np.shape(sub_output) + sub_output = np.transpose(np.reshape(sub_output, [b, 3, -1, h, w]), [0, 3, 4, 1, 2])[0] + score = np.max(sigmoid(sub_output[..., 4]), -1) + score = cv2.resize(score, (image.size[0], image.size[1])) + normed_score = (score * 255).astype('uint8') + mask = np.maximum(mask, normed_score) + + plt.imshow(mask, alpha=0.5, interpolation='nearest', cmap="jet") + + plt.axis('off') + plt.subplots_adjust(top=1, bottom=0, right=1, left=0, hspace=0, wspace=0) + plt.margins(0, 0) + plt.savefig(heatmap_save_path, dpi=200, bbox_inches='tight', pad_inches = -0.1) + print("Save to the " + heatmap_save_path) + plt.show() + + def convert_to_onnx(self, simplify, model_path): + import onnx + self.generate(onnx=True) + + im = torch.zeros(1, 3, *self.input_shape).to('cpu') # image size(1, 3, 512, 512) BCHW + input_layer_names = ["images"] + output_layer_names = ["output"] + + # Export the model + print(f'Starting export with onnx {onnx.__version__}.') + torch.onnx.export(self.net, + im, + f = model_path, + verbose = False, + opset_version = 12, + training = torch.onnx.TrainingMode.EVAL, + do_constant_folding = True, + input_names = input_layer_names, + output_names = output_layer_names, + dynamic_axes = None) + + # Checks + model_onnx = onnx.load(model_path) # load onnx model + onnx.checker.check_model(model_onnx) # check onnx model + + # Simplify onnx + if simplify: + import onnxsim + print(f'Simplifying with onnx-simplifier {onnxsim.__version__}.') + model_onnx, check = onnxsim.simplify( + model_onnx, + dynamic_input_shape=False, + input_shapes=None) + assert check, 'assert check failed' + onnx.save(model_onnx, model_path) + + print('Onnx model save as {}'.format(model_path)) + + def get_map_txt(self, image_id, image, class_names, map_out_path): + f = open(os.path.join(map_out_path, "detection-results/"+image_id+".txt"), "w", encoding='utf-8') + image_shape = np.array(np.shape(image)[0:2]) + #---------------------------------------------------------# + # 在这里将图像转换成RGB图像，防止灰度图在预测时报错。 + # 代码仅仅支持RGB图像的预测，所有其它类型的图像都会转化成RGB + #---------------------------------------------------------# + image = cvtColor(image) + #---------------------------------------------------------# + # 给图像增加灰条，实现不失真的resize + # 也可以直接resize进行识别 + #---------------------------------------------------------# + image_data = resize_image(image, (self.input_shape[1], self.input_shape[0]), self.letterbox_image) + #---------------------------------------------------------# + # 添加上batch_size维度 + #---------------------------------------------------------# + image_data = np.expand_dims(np.transpose(preprocess_input(np.array(image_data, dtype='float32')), (2, 0, 1)), 0) + + with torch.no_grad(): + images = torch.from_numpy(image_data) + if self.cuda: + images = images.cuda() + #---------------------------------------------------------# + # 将图像输入网络当中进行预测！ + #---------------------------------------------------------# + outputs = self.net(images) + outputs = self.bbox_util.decode_box(outputs) + #---------------------------------------------------------# + # 将预测框进行堆叠，然后进行非极大抑制 + #---------------------------------------------------------# + results = self.bbox_util.non_max_suppression(torch.cat(outputs, 1), self.num_classes, self.input_shape, + image_shape, self.letterbox_image, conf_thres = self.confidence, nms_thres = self.nms_iou) + + if results[0] is None: + return + + top_label = np.array(results[0][:, 6], dtype = 'int32') + top_conf = results[0][:, 4] * results[0][:, 5] + top_boxes = results[0][:, :4] + + for i, c in list(enumerate(top_label)): + predicted_class = self.class_names[int(c)] + box = top_boxes[i] + score = str(top_conf[i]) + + top, left, bottom, right = box + if predicted_class not in class_names: + continue + + f.write("%s %s %s %s %s %s\n" % (predicted_class, score[:6], str(int(left)), str(int(top)), str(int(right)),str(int(bottom)))) + + f.close() + return + +class YOLO_ONNX(object): + _defaults = { + #--------------------------------------------------------------------------# + # 使用自己训练好的模型进行预测一定要修改onnx_path和classes_path！ + # onnx_path指向logs文件夹下的权值文件，classes_path指向model_data下的txt + # + # 训练好后logs文件夹下存在多个权值文件，选择验证集损失较低的即可。 + # 验证集损失较低不代表mAP较高，仅代表该权值在验证集上泛化性能较好。 + # 如果出现shape不匹配，同时要注意训练时的onnx_path和classes_path参数的修改 + #--------------------------------------------------------------------------# + "onnx_path" : 'model_data/models.onnx', + "classes_path" : 'model_data/coco_classes.txt', + #---------------------------------------------------------------------# + # anchors_path代表先验框对应的txt文件，一般不修改。 + # anchors_mask用于帮助代码找到对应的先验框，一般不修改。 + #---------------------------------------------------------------------# + "anchors_path" : 'model_data/yolo_anchors.txt', + "anchors_mask" : [[6, 7, 8], [3, 4, 5], [0, 1, 2]], + #---------------------------------------------------------------------# + # 输入图片的大小，必须为32的倍数。 + #---------------------------------------------------------------------# + "input_shape" : [640, 640], + #---------------------------------------------------------------------# + # 只有得分大于置信度的预测框会被保留下来 + #---------------------------------------------------------------------# + "confidence" : 0.5, + #---------------------------------------------------------------------# + # 非极大抑制所用到的nms_iou大小 + #---------------------------------------------------------------------# + "nms_iou" : 0.3, + #---------------------------------------------------------------------# + # 该变量用于控制是否使用letterbox_image对输入图像进行不失真的resize， + # 在多次测试后，发现关闭letterbox_image直接resize的效果更好 + #---------------------------------------------------------------------# + "letterbox_image" : True + } + + @classmethod + def get_defaults(cls, n): + if n in cls._defaults: + return cls._defaults[n] + else: + return "Unrecognized attribute name '" + n + "'" + + #---------------------------------------------------# + # 初始化YOLO + #---------------------------------------------------# + def __init__(self, **kwargs): + self.__dict__.update(self._defaults) + for name, value in kwargs.items(): + setattr(self, name, value) + self._defaults[name] = value + + import onnxruntime + self.onnx_session = onnxruntime.InferenceSession(self.onnx_path) + # 获得所有的输入node + self.input_name = self.get_input_name() + # 获得所有的输出node + self.output_name = self.get_output_name() + + #---------------------------------------------------# + # 获得种类和先验框的数量 + #---------------------------------------------------# + self.class_names, self.num_classes = self.get_classes(self.classes_path) + self.anchors, self.num_anchors = self.get_anchors(self.anchors_path) + self.bbox_util = DecodeBoxNP(self.anchors, self.num_classes, (self.input_shape[0], self.input_shape[1]), self.anchors_mask) + + #---------------------------------------------------# + # 画框设置不同的颜色 + #---------------------------------------------------# + hsv_tuples = [(x / self.num_classes, 1., 1.) for x in range(self.num_classes)] + self.colors = list(map(lambda x: colorsys.hsv_to_rgb(*x), hsv_tuples)) + self.colors = list(map(lambda x: (int(x[0] * 255), int(x[1] * 255), int(x[2] * 255)), self.colors)) + + show_config(**self._defaults) + + def get_classes(self, classes_path): + with open(classes_path, encoding='utf-8') as f: + class_names = f.readlines() + class_names = [c.strip() for c in class_names] + return class_names, len(class_names) + + def get_anchors(self, anchors_path): + '''loads the anchors from a file''' + with open(anchors_path, encoding='utf-8') as f: + anchors = f.readline() + anchors = [float(x) for x in anchors.split(',')] + anchors = np.array(anchors).reshape(-1, 2) + return anchors, len(anchors) + + def get_input_name(self): + # 获得所有的输入node + input_name=[] + for node in self.onnx_session.get_inputs(): + input_name.append(node.name) + return input_name + + def get_output_name(self): + # 获得所有的输出node + output_name=[] + for node in self.onnx_session.get_outputs(): + output_name.append(node.name) + return output_name + + def get_input_feed(self,image_tensor): + # 利用input_name获得输入的tensor + input_feed={} + for name in self.input_name: + input_feed[name]=image_tensor + return input_feed + + #---------------------------------------------------# + # 对输入图像进行resize + #---------------------------------------------------# + def resize_image(self, image, size, letterbox_image, mode='PIL'): + if mode == 'PIL': + iw, ih = image.size + w, h = size + + if letterbox_image: + scale = min(w/iw, h/ih) + nw = int(iw*scale) + nh = int(ih*scale) + + image = image.resize((nw,nh), Image.BICUBIC) + new_image = Image.new('RGB', size, (128,128,128)) + new_image.paste(image, ((w-nw)//2, (h-nh)//2)) + else: + new_image = image.resize((w, h), Image.BICUBIC) + else: + image = np.array(image) + if letterbox_image: + # 获得现在的shape + shape = np.shape(image)[:2] + # 获得输出的shape + if isinstance(size, int): + size = (size, size) + + # 计算缩放的比例 + r = min(size[0] / shape[0], size[1] / shape[1]) + + # 计算缩放后图片的高宽 + new_unpad = int(round(shape[1] * r)), int(round(shape[0] * r)) + dw, dh = size[1] - new_unpad[0], size[0] - new_unpad[1] + + # 除以2以padding到两边 + dw /= 2 + dh /= 2 + + # 对图像进行resize + if shape[::-1] != new_unpad: # resize + image = cv2.resize(image, new_unpad, interpolation=cv2.INTER_LINEAR) + top, bottom = int(round(dh - 0.1)), int(round(dh + 0.1)) + left, right = int(round(dw - 0.1)), int(round(dw + 0.1)) + + new_image = cv2.copyMakeBorder(image, top, bottom, left, right, cv2.BORDER_CONSTANT, value=(128, 128, 128)) # add border + else: + new_image = cv2.resize(image, (w, h)) + + return new_image + + def detect_image(self, image): + image_shape = np.array(np.shape(image)[0:2]) + #---------------------------------------------------------# + # 在这里将图像转换成RGB图像，防止灰度图在预测时报错。 + # 代码仅仅支持RGB图像的预测，所有其它类型的图像都会转化成RGB + #---------------------------------------------------------# + image = cvtColor(image) + + image_data = self.resize_image(image, self.input_shape, True) + #---------------------------------------------------------# + # 添加上batch_size维度 + # h, w, 3 => 3, h, w => 1, 3, h, w + #---------------------------------------------------------# + image_data = np.expand_dims(np.transpose(preprocess_input(np.array(image_data, dtype='float32')), (2, 0, 1)), 0) + + input_feed = self.get_input_feed(image_data) + outputs = self.onnx_session.run(output_names=self.output_name, input_feed=input_feed) + + feature_map_shape = [[int(j / (2 ** (i + 3))) for j in self.input_shape] for i in range(len(self.anchors_mask))][::-1] + for i in range(len(self.anchors_mask)): + outputs[i] = np.reshape(outputs[i], (1, len(self.anchors_mask[i]) * (5 + self.num_classes), feature_map_shape[i][0], feature_map_shape[i][1])) + + outputs = self.bbox_util.decode_box(outputs) + #---------------------------------------------------------# + # 将预测框进行堆叠，然后进行非极大抑制 + #---------------------------------------------------------# + results = self.bbox_util.non_max_suppression(np.concatenate(outputs, 1), self.num_classes, self.input_shape, + image_shape, self.letterbox_image, conf_thres = self.confidence, nms_thres = self.nms_iou) + + if results[0] is None: + return image + + top_label = np.array(results[0][:, 6], dtype = 'int32') + top_conf = results[0][:, 4] * results[0][:, 5] + top_boxes = results[0][:, :4] + + #---------------------------------------------------------# + # 设置字体与边框厚度 + #---------------------------------------------------------# + font = ImageFont.truetype(font='model_data/simhei.ttf', size=np.floor(3e-2 * image.size[1] + 0.5).astype('int32')) + thickness = int(max((image.size[0] + image.size[1]) // np.mean(self.input_shape), 1)) + + #---------------------------------------------------------# + # 图像绘制 + #---------------------------------------------------------# + for i, c in list(enumerate(top_label)): + predicted_class = self.class_names[int(c)] + box = top_boxes[i] + score = top_conf[i] + + top, left, bottom, right = box + + top = max(0, np.floor(top).astype('int32')) + left = max(0, np.floor(left).astype('int32')) + bottom = min(image.size[1], np.floor(bottom).astype('int32')) + right = min(image.size[0], np.floor(right).astype('int32')) + + label = '{} {:.2f}'.format(predicted_class, score) + draw = ImageDraw.Draw(image) + label_size = draw.textsize(label, font) + label = label.encode('utf-8') + print(label, top, left, bottom, right) + + if top - label_size[1] >= 0: + text_origin = np.array([left, top - label_size[1]]) + else: + text_origin = np.array([left, top + 1]) + + for i in range(thickness): + draw.rectangle([left + i, top + i, right - i, bottom - i], outline=self.colors[c]) + draw.rectangle([tuple(text_origin), tuple(text_origin + label_size)], fill=self.colors[c]) + draw.text(text_origin, str(label,'UTF-8'), fill=(0, 0, 0), font=font) + del draw + + return image \ No newline at end of file