我对代码进行了一些优化和改进

datadeal/deepfasion2_to_yolo.py

@@ -17,9 +17,9 @@ def listPathAllfiles(dirname):
 
 
 if __name__ == '__main__':
-    annos_path = r"E:\06服饰\Deepfashion2\train\train\annos"  # 改成需要路径
-    image_path = r"E:\06服饰\Deepfashion2\train\train\images"  # 改成需要路径
-    labels_path = r"E:\06服饰\Deepfashion2\train\train\labels"  # 改成需要路径
+    annos_path = r"F:\Deepfashion2\validation\annos"  # 改成需要路径
+    image_path = r"F:\Deepfashion2\validation\images"  # 改成需要路径
+    labels_path = r"F:\Deepfashion2\validation\labels"  # 改成需要路径
 
     num_images = len(os.listdir(annos_path))
 
@@ -37,7 +37,7 @@ if __name__ == '__main__':
                 if i == 'source' or i == 'pair_id':
                     continue
                 else:
-                    box = temp[i]['bounding_box']
+                    box = temp[i]['segmentation']
                     x_1 = round((box[0] + box[2]) / 2 / width, 6)
                     y_1 = round((box[1] + box[3]) / 2 / height, 6)
                     w = round((box[2] - box[0]) / width, 6)

datadeal/deepfasion2_to_yolo_seg.py

+# coding:utf-8
+import json
+import os
+import os.path
+
+from PIL import Image
+from tqdm import tqdm
+
+
+def listPathAllfiles(dirname):
+    result = []
+    for maindir, subdir, file_name_list in os.walk(dirname):
+        for filename in file_name_list:
+            apath = os.path.join(maindir, filename)
+            result.append(apath)
+    return result
+
+
+if __name__ == '__main__':
+    annos_path = r"F:\Deepfashion2\validation\annos"  # 改成需要路径
+    image_path = r"F:\Deepfashion2\validation\images"  # 改成需要路径
+    labels_path = r"F:\Deepfashion2\validation\labels"  # 改成需要路径
+
+    num_images = len(os.listdir(annos_path))
+
+    for num in tqdm(range(1, num_images + 1)):
+        json_name = os.path.join(annos_path, str(num).zfill(6) + '.json')
+        image_name = os.path.join(image_path, str(num).zfill(6) + '.jpg')
+        txtfile = os.path.join(labels_path, str(num).zfill(6) + '.txt')
+        imag = Image.open(image_name)
+        width, height = imag.size
+
+        res = []
+        with open(json_name, 'r') as f:
+            temp = json.loads(f.read())
+            for i in temp:
+                if i == 'source' or i == 'pair_id':
+                    continue
+                else:
+                    box = temp[i]['segmentation']
+                    category_id = int(temp[i]['category_id'] - 1)
+                    list_nums=[]
+                    list_nums.append(str(category_id))
+
+
+
+
+                    for index in range(len(box)):
+                        for indexj in range(len(box[index])):
+                            list_nums.append(str(box[index][indexj]))
+                    res.append(" ".join(list_nums))
+                    list_nums.clear()
+
+        open(txtfile, "w").write("\n".join(res))
+

datadeal/step2-fen_trainval.py

@@ -2,7 +2,7 @@ import os
 import random
 import sys
 
-root_path = r'E:/byc/mythree_traindatasets' #要改
+root_path = r'F:/Deepfashion2/fours_seg/' #要改
 label_path = root_path + r'/labels' #txt格式标签文件的位置（要改）
 fenpei_xinxi_path = root_path + r'/fenpei' #分配信息保存位置
 
@@ -10,7 +10,7 @@ if not os.path.exists(fenpei_xinxi_path):
     os.makedirs(fenpei_xinxi_path)
 
 train_test_percent = 1.0  # (训练集+验证集)/(训练集+验证集+测试集)=1即没有测试集
-train_valid_percent = 1.0  # 训练集/(训练集+验证集)
+train_valid_percent = 0.8 # 训练集/(训练集+验证集)
 
 total_txt = os.listdir(label_path)
 num = len(total_txt)

datadeal/step3-chonggou.py

@@ -64,7 +64,7 @@ def chonggou(root_path, fenpei_path,img_path,label_path):
         shutil.copy(label_path + str(label_txt_cg_valid[q]), new_dataset_validl)
 
 if __name__ == "__main__":
-    root_path = r'E:/byc/mythree_traindatasets'  #根目录
+    root_path = r'F:/Deepfashion2/fours_seg/'  #根目录
     fenpei_path = root_path + r'/fenpei'   #step2输出的分配信息（6个txt文件）
     img_path = root_path + '\\images\\' #原图位置
     label_path = root_path + '\\labels\\'#原txt标签位置

demo1.py

 from ultralytics import RTDETR
 from ultralytics import YOLO
 # Load a model
-model = RTDETR("E:/fourworkplace/yolov8v4/ultralytics/cfg/models/v8/yolov8-swinV2-RTViT.yaml")  # build a new model from scratch
+model = RTDETR("D:/BProjectAll/workplace/yolov8V4/ultralytics/cfg/models/v8/yolov8-swinV2-RTViT.yaml")  # build a new model from scratch
 model.info()
 
-model1=RTDETR("E:/fourworkplace/yolov8v4/ultralytics/cfg/models/v8/yolov8-swinV2-RT.yaml")
+model1=RTDETR("D:/BProjectAll/workplace/yolov8V4/ultralytics/cfg/models/v8/yolov8-swinV2-RT.yaml")
 model1.info()
 
-model2=RTDETR("E:/fourworkplace/yolov8v4/ultralytics/cfg/models/v5/yolov5-swinV2-RTViT.yaml")
+model2=RTDETR("D:/BProjectAll/workplace/yolov8V4/ultralytics/cfg/models/v5/yolov5-swinV2-RTViT.yaml")
 model2.info()
 
 
 
-model3=RTDETR("E:/fourworkplace/yolov8v4/ultralytics/cfg/models/v3/yolov3-swinV2-RTViT.yaml")
+model3=RTDETR("D:/BProjectAll/workplace/yolov8V4/ultralytics/cfg/models/v3/yolov3-swinV2-RTViT.yaml")
 model3.info()
 
 

myfourtrain.py

 from ultralytics import RTDETR
 from ultralytics import YOLO
 
-epoch_set=100
+epoch_set=300
 workers_set=14
+model_yaml_path=""
+data_yaml_path=""
 
 #开始进行基础的YOLOv5原版训练
-model = YOLO("E:/fourworkplace/yolov8v4/ultralytics/cfg/models/v5/yolov5.yaml")  # build a new model from scratch
-model.train(data="E:/fourworkplace/yolov8v4/ultralytics/cfg/datasets/yolov8_four.yaml",
+model = YOLO(model_yaml_path+"v5/yolov5.yaml")  # build a new model from scratch
+model.train(data=data_yaml_path+"datasets/yolov8_four.yaml",
     pretrained=True,epochs=epoch_set,batch=64,patience=200,resume=True,workers=workers_set)
 
 #本文的改进方案在YOLOv5上产生的效果CNN+SiwnV2+RTViT
-model = YOLO("E:/fourworkplace/yolov8v4/ultralytics/cfg/models/v5/yolov5-swinV2-RTViT.yaml")  # build a new model from scratch
-model.train(data="E:/fourworkplace/yolov8v4/ultralytics/cfg/datasets/yolov8_four.yaml",
+model = RTDETR(model_yaml_path+"v5/yolov5-swinV2-RTViT.yaml")  # build a new model from scratch
+model.train(data=data_yaml_path+"datasets/yolov8_four.yaml",
     pretrained=True,epochs=epoch_set,batch=64,patience=200,resume=True,workers=workers_set)
 
 #开始进行基础的YOLOv8原版训练
-model = YOLO("E:/fourworkplace/yolov8v4/ultralytics/cfg/models/v8/yolov8.yaml")  # build a new model from scratch
-model.train(data="E:/fourworkplace/yolov8v4/ultralytics/cfg/datasets/yolov8_four.yaml",
+model = YOLO(model_yaml_path+"v8/yolov8.yaml")  # build a new model from scratch
+model.train(data=data_yaml_path+"datasets/yolov8_four.yaml",
     pretrained=True,epochs=epoch_set,batch=64,patience=200,resume=True,workers=workers_set)
 
 #训练YOLOv8+Swin替换(使用CNN+SwinV2进行训练)
-model = YOLO("E:/fourworkplace/yolov8v4/ultralytics/cfg/models/v8/yolov8-swinV2-detect.yaml")  # build a new model from scratch
-model.train(data="E:/fourworkplace/yolov8v4/ultralytics/cfg/datasets/yolov8_four.yaml",
+model = YOLO(model_yaml_path+"v8/yolov8-swinV2-detect.yaml")  # build a new model from scratch
+model.train(data=data_yaml_path+"datasets/yolov8_four.yaml",
     pretrained=True,epochs=epoch_set,batch=64,patience=200,resume=True,workers=workers_set)
 
 
 #训练YOLOv8+Swin替换(使用CNN+SwinV2+RT进行训练)
-model = YOLO("E:/fourworkplace/yolov8v4/ultralytics/cfg/models/v8/yolov8-swinV2-RT.yaml")  # build a new model from scratch
-model.train(data="E:/fourworkplace/yolov8v4/ultralytics/cfg/datasets/yolov8_four.yaml",
+model = RTDETR(model_yaml_path+"v8/yolov8-swinV2-RT.yaml")  # build a new model from scratch
+model.train(data=data_yaml_path+"datasets/yolov8_four.yaml",
+    pretrained=True,epochs=epoch_set,batch=64,patience=200,resume=True,workers=workers_set)
+
+#训练YOLOv8-rtdetr进行训练
+model = RTDETR(model_yaml_path+"v8/yolov8-rtdetr.yaml")  # build a new model from scratch
+model.train(data=data_yaml_path+"datasets/yolov8_four.yaml",
     pretrained=True,epochs=epoch_set,batch=64,patience=200,resume=True,workers=workers_set)
 
 #训练YOLOv8+Swin替换(使用CNN+SwinV2+RTViT进行训练)
-model = YOLO("E:/fourworkplace/yolov8v4/ultralytics/cfg/models/v8/yolov8-swinV2-RTViT.yaml")  # build a new model from scratch
-model.train(data="E:/fourworkplace/yolov8v4/ultralytics/cfg/datasets/yolov8_four.yaml",
+model = RTDETR(model_yaml_path+"v8/yolov8-swinV2-RTViT.yaml")  # build a new model from scratch
+model.train(data=data_yaml_path+"datasets/yolov8_four.yaml",
+    pretrained=True,epochs=epoch_set,batch=64,patience=200,resume=True,workers=workers_set)
+
+#使用YOLOv3进行训练,并把我们的改进方法加到YOLOv3上面以此突出改进方案的优势。
+#训练官网原版本的YOLOv3,考虑到YOLOV3的特殊性batch要特别设置，不要太大，设置为32吧
+model = YOLO(model_yaml_path+"v3/yolov3.yaml")  # build a new model from scratch
+model.train(data=data_yaml_path+"datasets/yolov8_four.yaml",
     pretrained=True,epochs=epoch_set,batch=64,patience=200,resume=True,workers=workers_set)
 
 #使用YOLOv3进行训练,并把我们的改进方法加到YOLOv3上面以此突出改进方案的优势。
+#训练官网原版本的YOLOv3+CNN+Swin+RTViT,考虑到YOLOV3的特殊性batch要特别设置，不要太大，设置为32吧
+model = RTDETR(model_yaml_path+"v3/yolov3-swinV2-RTViT.yaml")  # build a new model from scratch
+model.train(data=data_yaml_path+"datasets/yolov8_four.yaml",
+    pretrained=True,epochs=epoch_set,batch=64,patience=200,resume=True,workers=workers_set)
 
+#下面开始考虑对图像分割问题的处理方案
+model = YOLO(model_yaml_path+"v8/yolov8-seg.yaml")  # build a new model from scratch
+model.train(data=data_yaml_path+"datasets/yolov8_seg_four.yaml",
+    pretrained=True,epochs=epoch_set,batch=64,patience=200,resume=True,workers=workers_set)
 
 #下面开始考虑对图像分割问题的处理方案
+model = YOLO(model_yaml_path+"v8/yolov8-seg-swinV2.yaml")  # build a new model from scratch
+model.train(data=data_yaml_path+"datasets/yolov8_seg_four.yaml",
+    pretrained=True,epochs=epoch_set,batch=64,patience=200,resume=True,workers=workers_set)
\ No newline at end of file

ultralytics/cfg/models/v8/yolov8-seg-swinV2.yaml

+# Ultralytics YOLO 🚀, AGPL-3.0 license
+# YOLOv8-seg instance segmentation model. For Usage examples see https://docs.ultralytics.com/tasks/segment
+
+# Parameters
+nc: 80  # number of classes
+scales: # model compound scaling constants, i.e. 'model=yolov8n-seg.yaml' will call yolov8-seg.yaml with scale 'n'
+  # [depth, width, max_channels]
+  n: [0.33, 0.25, 1024]
+  s: [0.33, 0.50, 1024]
+  m: [0.67, 0.75, 768]
+  l: [1.00, 1.00, 512]
+  x: [1.00, 1.25, 512]
+
+# YOLOv8.0n backbone
+backbone:
+  # [from, repeats, module, args]
+  - [-1, 1, Conv, [64, 3, 2]]  # 0-P1/2
+  - [-1, 1, Conv, [128, 3, 2]]  # 1-P2/4
+  - [-1, 3, SwinV2_CSPB, [128, True]]
+  - [-1, 1, Conv, [256, 3, 2]]  # 3-P3/8
+  - [-1, 6, SwinV2_CSPB, [256, True]]
+  - [-1, 1, Conv, [512, 3, 2]]  # 5-P4/16
+  - [-1, 6, SwinV2_CSPB, [512, True]]
+  - [-1, 1, Conv, [1024, 3, 2]]  # 7-P5/32
+  - [-1, 3, SwinV2_CSPB, [1024, True]]
+  - [-1, 1, SPPF, [1024, 5]]  # 9
+
+# YOLOv8.0n head
+head:
+  - [-1, 1, nn.Upsample, [None, 2, 'nearest']]
+  - [[-1, 6], 1, Concat, [1]]  # cat backbone P4
+  - [-1, 3, SwinV2_CSPB, [512]]  # 12
+
+  - [-1, 1, nn.Upsample, [None, 2, 'nearest']]
+  - [[-1, 4], 1, Concat, [1]]  # cat backbone P3
+  - [-1, 3, SwinV2_CSPB, [256]]  # 15 (P3/8-small)
+
+  - [-1, 1, Conv, [256, 3, 2]]
+  - [[-1, 12], 1, Concat, [1]]  # cat head P4
+  - [-1, 3, SwinV2_CSPB, [512]]  # 18 (P4/16-medium)
+
+  - [-1, 1, Conv, [512, 3, 2]]
+  - [[-1, 9], 1, Concat, [1]]  # cat head P5
+  - [-1, 3, SwinV2_CSPB, [1024]]  # 21 (P5/32-large)
+
+  - [[15, 18, 21], 1, Segment, [nc, 32, 256]]  # Segment(P3, P4, P5)