fix

Image/AlexNet/code/train.py

@@ -7,14 +7,14 @@ from utils.parse_args import parse_args
 from model import AlexNet
 #args.train_type #0 for normal train, 1 for data aug train,2 for back door train
 
-def main(train_type):
+def main():
     # 解析命令行参数
     args = parse_args()
     # 创建模型
     model = AlexNet()
     if args.train_type == '0':
         # 获取数据加载器
-        trainloader, testloader = get_cifar10_dataloaders(batch_size=args.batch_size)
+        trainloader, testloader = get_cifar10_dataloaders(batch_size=args.batch_size, local_dataset_path=args.dataset_path)
         # 训练模型
         train_model(
             model=model,
@@ -29,11 +29,13 @@ def main(train_type):
     elif args.train_type == '1':
         train_model_data_augmentation(model, epochs=args.epochs, lr=args.lr, device=f'cuda:{args.gpu}', 
                                 save_dir='../model', model_name='alexnet',
-                                batch_size=args.batch_size, num_workers=args.num_workers)
+                                batch_size=args.batch_size, num_workers=args.num_workers,
+                                local_dataset_path=args.dataset_path)
     elif args.train_type == '2':
         train_model_backdoor(model, poison_ratio=0.1, target_label=0, epochs=args.epochs, lr=args.lr,
                         device=f'cuda:{args.gpu}', save_dir='../model', model_name='alexnet',
-                        batch_size=args.batch_size, num_workers=args.num_workers)
+                        batch_size=args.batch_size, num_workers=args.num_workers,
+                        local_dataset_path=args.dataset_path)
     
 if __name__ == '__main__': 
     main()

Image/DenseNet/code/train.py

@@ -15,7 +15,7 @@ def main():
     
     if args.train_type == '0':
         # 获取数据加载器
-        trainloader, testloader = get_cifar10_dataloaders(batch_size=args.batch_size)
+        trainloader, testloader = get_cifar10_dataloaders(batch_size=args.batch_size, local_dataset_path=args.dataset_path)
         # 训练模型
         train_model(
             model=model,
@@ -37,7 +37,8 @@ def main():
             save_dir='../model', 
             model_name='densenet',
             batch_size=args.batch_size, 
-            num_workers=args.num_workers
+            num_workers=args.num_workers,
+            local_dataset_path=args.dataset_path
         )
     elif args.train_type == '2':
         train_model_backdoor(
@@ -50,7 +51,8 @@ def main():
             save_dir='../model', 
             model_name='densenet',
             batch_size=args.batch_size, 
-            num_workers=args.num_workers
+            num_workers=args.num_workers,
+            local_dataset_path=args.dataset_path 
         )
 
 if __name__ == '__main__':

Image/EfficientNet/code/train.py

@@ -15,7 +15,7 @@ def main():
     
     if args.train_type == '0':
         # 获取数据加载器
-        trainloader, testloader = get_cifar10_dataloaders(batch_size=args.batch_size)
+        trainloader, testloader = get_cifar10_dataloaders(batch_size=args.batch_size, local_dataset_path=args.dataset_path)
         # 训练模型
         train_model(
             model=model,
@@ -37,7 +37,8 @@ def main():
             save_dir='../model', 
             model_name='efficientnet',
             batch_size=args.batch_size, 
-            num_workers=args.num_workers
+            num_workers=args.num_workers,
+            local_dataset_path=args.dataset_path
         )
     elif args.train_type == '2':
         train_model_backdoor(
@@ -50,7 +51,8 @@ def main():
             save_dir='../model', 
             model_name='efficientnet',
             batch_size=args.batch_size, 
-            num_workers=args.num_workers
+            num_workers=args.num_workers,
+            local_dataset_path=args.dataset_path
         )
 
 if __name__ == '__main__':

Image/GoogLeNet/code/train.py

@@ -15,7 +15,7 @@ def main():
     
     if args.train_type == '0':
         # 获取数据加载器
-        trainloader, testloader = get_cifar10_dataloaders(batch_size=args.batch_size)
+        trainloader, testloader = get_cifar10_dataloaders(batch_size=args.batch_size, local_dataset_path=args.dataset_path)
         # 训练模型
         train_model(
             model=model,
@@ -37,7 +37,8 @@ def main():
             save_dir='../model', 
             model_name='googlenet',
             batch_size=args.batch_size, 
-            num_workers=args.num_workers
+            num_workers=args.num_workers,
+            local_dataset_path=args.dataset_path
         )
     elif args.train_type == '2':
         train_model_backdoor(
@@ -50,7 +51,8 @@ def main():
             save_dir='../model', 
             model_name='googlenet',
             batch_size=args.batch_size, 
-            num_workers=args.num_workers
+            num_workers=args.num_workers,
+            local_dataset_path=args.dataset_path
         )
 
 if __name__ == '__main__':

Image/LeNet5/code/train.py

@@ -15,7 +15,7 @@ def main():
     
     if args.train_type == '0':
         # 获取数据加载器
-        trainloader, testloader = get_cifar10_dataloaders(batch_size=args.batch_size)
+        trainloader, testloader = get_cifar10_dataloaders(batch_size=args.batch_size, local_dataset_path=args.dataset_path)
         # 训练模型
         train_model(
             model=model,
@@ -37,7 +37,8 @@ def main():
             save_dir='../model', 
             model_name='lenet5',
             batch_size=args.batch_size, 
-            num_workers=args.num_workers
+            num_workers=args.num_workers,
+            local_dataset_path=args.dataset_path
         )
     elif args.train_type == '2':
         train_model_backdoor(
@@ -50,7 +51,8 @@ def main():
             save_dir='../model', 
             model_name='lenet5',
             batch_size=args.batch_size, 
-            num_workers=args.num_workers
+            num_workers=args.num_workers,
+            local_dataset_path=args.dataset_path
         )
 
 if __name__ == '__main__':

Image/MobileNetv1/code/train.py

@@ -15,7 +15,7 @@ def main():
     
     if args.train_type == '0':
         # 获取数据加载器
-        trainloader, testloader = get_cifar10_dataloaders(batch_size=args.batch_size)
+        trainloader, testloader = get_cifar10_dataloaders(batch_size=args.batch_size, local_dataset_path=args.dataset_path)
         # 训练模型
         train_model(
             model=model,
@@ -37,7 +37,8 @@ def main():
             save_dir='../model', 
             model_name='mobilenetv1',
             batch_size=args.batch_size, 
-            num_workers=args.num_workers
+            num_workers=args.num_workers,
+            local_dataset_path=args.dataset_path
         )
     elif args.train_type == '2':
         train_model_backdoor(
@@ -50,7 +51,8 @@ def main():
             save_dir='../model', 
             model_name='mobilenetv1',
             batch_size=args.batch_size, 
-            num_workers=args.num_workers
+            num_workers=args.num_workers,
+            local_dataset_path=args.dataset_path
         )
 
 if __name__ == '__main__':

Image/MobileNetv2/code/train.py

@@ -15,7 +15,7 @@ def main():
     
     if args.train_type == '0':
         # 获取数据加载器
-        trainloader, testloader = get_cifar10_dataloaders(batch_size=args.batch_size)
+        trainloader, testloader = get_cifar10_dataloaders(batch_size=args.batch_size, local_dataset_path=args.dataset_path)
         # 训练模型
         train_model(
             model=model,
@@ -37,7 +37,8 @@ def main():
             save_dir='../model', 
             model_name='mobilenetv2',
             batch_size=args.batch_size, 
-            num_workers=args.num_workers
+            num_workers=args.num_workers,
+            local_dataset_path=args.dataset_path
         )
     elif args.train_type == '2':
         train_model_backdoor(
@@ -50,7 +51,8 @@ def main():
             save_dir='../model', 
             model_name='mobilenetv2',
             batch_size=args.batch_size, 
-            num_workers=args.num_workers
+            num_workers=args.num_workers,
+            local_dataset_path=args.dataset_path
         )
 
 if __name__ == '__main__':

Image/MobileNetv3/code/train.py

@@ -15,7 +15,7 @@ def main():
     
     if args.train_type == '0':
         # 获取数据加载器
-        trainloader, testloader = get_cifar10_dataloaders(batch_size=args.batch_size)
+        trainloader, testloader = get_cifar10_dataloaders(batch_size=args.batch_size, local_dataset_path=args.dataset_path)
         # 训练模型
         train_model(
             model=model,
@@ -37,7 +37,8 @@ def main():
             save_dir='../model', 
             model_name='mobilenetv3',
             batch_size=args.batch_size, 
-            num_workers=args.num_workers
+            num_workers=args.num_workers,
+            local_dataset_path=args.dataset_path
         )
     elif args.train_type == '2':
         train_model_backdoor(
@@ -50,7 +51,8 @@ def main():
             save_dir='../model', 
             model_name='mobilenetv3',
             batch_size=args.batch_size, 
-            num_workers=args.num_workers
+            num_workers=args.num_workers,
+            local_dataset_path=args.dataset_path
         )
 
 if __name__ == '__main__':

Image/ResNet/code/train.py

@@ -15,7 +15,7 @@ def main():
     
     if args.train_type == '0':
         # 获取数据加载器
-        trainloader, testloader = get_cifar10_dataloaders(batch_size=args.batch_size)
+        trainloader, testloader = get_cifar10_dataloaders(batch_size=args.batch_size, local_dataset_path=args.dataset_path)
         # 训练模型
         train_model(
             model=model,
@@ -37,7 +37,8 @@ def main():
             save_dir='../model', 
             model_name='resnet',
             batch_size=args.batch_size, 
-            num_workers=args.num_workers
+            num_workers=args.num_workers,
+            local_dataset_path=args.dataset_path
         )
     elif args.train_type == '2':
         train_model_backdoor(
@@ -50,7 +51,8 @@ def main():
             save_dir='../model', 
             model_name='resnet',
             batch_size=args.batch_size, 
-            num_workers=args.num_workers
+            num_workers=args.num_workers,
+            local_dataset_path=args.dataset_path
         )
 
 if __name__ == '__main__':

Image/SENet/code/train.py

@@ -15,7 +15,7 @@ def main():
     
     if args.train_type == '0':
         # 获取数据加载器
-        trainloader, testloader = get_cifar10_dataloaders(batch_size=args.batch_size)
+        trainloader, testloader = get_cifar10_dataloaders(batch_size=args.batch_size, local_dataset_path=args.dataset_path)
         # 训练模型
         train_model(
             model=model,
@@ -37,7 +37,8 @@ def main():
             save_dir='../model', 
             model_name='senet',
             batch_size=args.batch_size, 
-            num_workers=args.num_workers
+            num_workers=args.num_workers,
+            local_dataset_path=args.dataset_path
         )
     elif args.train_type == '2':
         train_model_backdoor(
@@ -50,7 +51,8 @@ def main():
             save_dir='../model', 
             model_name='senet',
             batch_size=args.batch_size, 
-            num_workers=args.num_workers
+            num_workers=args.num_workers,
+            local_dataset_path=args.dataset_path
         )
 
 if __name__ == '__main__':

Image/ShuffleNet/code/train.py

@@ -15,7 +15,7 @@ def main():
     
     if args.train_type == '0':
         # 获取数据加载器
-        trainloader, testloader = get_cifar10_dataloaders(batch_size=args.batch_size)
+        trainloader, testloader = get_cifar10_dataloaders(batch_size=args.batch_size, local_dataset_path=args.dataset_path)
         # 训练模型
         train_model(
             model=model,
@@ -37,7 +37,8 @@ def main():
             save_dir='../model', 
             model_name='shufflenet',
             batch_size=args.batch_size, 
-            num_workers=args.num_workers
+            num_workers=args.num_workers,
+            local_dataset_path=args.dataset_path
         )
     elif args.train_type == '2':
         train_model_backdoor(
@@ -50,7 +51,8 @@ def main():
             save_dir='../model', 
             model_name='shufflenet',
             batch_size=args.batch_size, 
-            num_workers=args.num_workers
+            num_workers=args.num_workers,
+            local_dataset_path=args.dataset_path
         )
 
 if __name__ == '__main__':

Image/ShuffleNetv2/code/train.py

@@ -15,7 +15,7 @@ def main():
     
     if args.train_type == '0':
         # 获取数据加载器
-        trainloader, testloader = get_cifar10_dataloaders(batch_size=args.batch_size)
+        trainloader, testloader = get_cifar10_dataloaders(batch_size=args.batch_size, local_dataset_path=args.dataset_path)
         # 训练模型
         train_model(
             model=model,
@@ -37,7 +37,8 @@ def main():
             save_dir='../model', 
             model_name='shufflenetv2',
             batch_size=args.batch_size, 
-            num_workers=args.num_workers
+            num_workers=args.num_workers,
+            local_dataset_path=args.dataset_path
         )
     elif args.train_type == '2':
         train_model_backdoor(
@@ -50,7 +51,8 @@ def main():
             save_dir='../model', 
             model_name='shufflenetv2',
             batch_size=args.batch_size, 
-            num_workers=args.num_workers
+            num_workers=args.num_workers,
+            local_dataset_path=args.dataset_path
         )
 
 if __name__ == '__main__':

Image/SwinTransformer/code/train.py

@@ -15,7 +15,7 @@ def main():
     
     if args.train_type == '0':
         # 获取数据加载器
-        trainloader, testloader = get_cifar10_dataloaders(batch_size=args.batch_size)
+        trainloader, testloader = get_cifar10_dataloaders(batch_size=args.batch_size, local_dataset_path=args.dataset_path)
         # 训练模型
         train_model(
             model=model,
@@ -37,7 +37,8 @@ def main():
             save_dir='../model', 
             model_name='swintransformer',
             batch_size=args.batch_size, 
-            num_workers=args.num_workers
+            num_workers=args.num_workers,
+            local_dataset_path=args.dataset_path
         )
     elif args.train_type == '2':
         train_model_backdoor(
@@ -50,7 +51,8 @@ def main():
             save_dir='../model', 
             model_name='swintransformer',
             batch_size=args.batch_size, 
-            num_workers=args.num_workers
+            num_workers=args.num_workers,
+            local_dataset_path=args.dataset_path
         )
 
 if __name__ == '__main__':

Image/VGG/code/train.py

@@ -15,7 +15,7 @@ def main():
     
     if args.train_type == '0':
         # 获取数据加载器
-        trainloader, testloader = get_cifar10_dataloaders(batch_size=args.batch_size)
+        trainloader, testloader = get_cifar10_dataloaders(batch_size=args.batch_size, local_dataset_path=args.dataset_path)
         # 训练模型
         train_model(
             model=model,
@@ -37,7 +37,8 @@ def main():
             save_dir='../model', 
             model_name='vgg',
             batch_size=args.batch_size, 
-            num_workers=args.num_workers
+            num_workers=args.num_workers,
+            local_dataset_path=args.dataset_path
         )
     elif args.train_type == '2':
         train_model_backdoor(
@@ -50,7 +51,8 @@ def main():
             save_dir='../model', 
             model_name='vgg',
             batch_size=args.batch_size, 
-            num_workers=args.num_workers
+            num_workers=args.num_workers,
+            local_dataset_path=args.dataset_path
         )
 
 if __name__ == '__main__':

Image/ViT/code/train.py

@@ -15,7 +15,7 @@ def main():
     
     if args.train_type == '0':
         # 获取数据加载器
-        trainloader, testloader = get_cifar10_dataloaders(batch_size=args.batch_size)
+        trainloader, testloader = get_cifar10_dataloaders(batch_size=args.batch_size, local_dataset_path=args.dataset_path)
         # 训练模型
         train_model(
             model=model,
@@ -37,7 +37,8 @@ def main():
             save_dir='../model', 
             model_name='vit',
             batch_size=args.batch_size, 
-            num_workers=args.num_workers
+            num_workers=args.num_workers,
+            local_dataset_path=args.dataset_path
         )
     elif args.train_type == '2':
         train_model_backdoor(
@@ -50,7 +51,8 @@ def main():
             save_dir='../model', 
             model_name='vit',
             batch_size=args.batch_size, 
-            num_workers=args.num_workers
+            num_workers=args.num_workers,
+            local_dataset_path=args.dataset_path
         )
 
 if __name__ == '__main__':

Image/ZFNet/code/train.py

@@ -15,7 +15,7 @@ def main():
     
     if args.train_type == '0':
         # 获取数据加载器
-        trainloader, testloader = get_cifar10_dataloaders(batch_size=args.batch_size)
+        trainloader, testloader = get_cifar10_dataloaders(batch_size=args.batch_size, local_dataset_path=args.dataset_path)
         # 训练模型
         train_model(
             model=model,
@@ -37,7 +37,8 @@ def main():
             save_dir='../model', 
             model_name='zfnet',
             batch_size=args.batch_size, 
-            num_workers=args.num_workers
+            num_workers=args.num_workers,
+            local_dataset_path=args.dataset_path
         )
     elif args.train_type == '2':
         train_model_backdoor(
@@ -50,7 +51,8 @@ def main():
             save_dir='../model', 
             model_name='zfnet',
             batch_size=args.batch_size, 
-            num_workers=args.num_workers
+            num_workers=args.num_workers,
+            local_dataset_path=args.dataset_path
         )
 
 if __name__ == '__main__':

Image/run_all_models.py

@@ -0,0 +1,59 @@
+import os
+import subprocess
+from pathlib import Path
+
+# 模型列表
+models = [
+    'AlexNet', 'DenseNet', 'EfficientNet', 'GoogLeNet', 'LeNet5',
+    'MobileNetv1', 'MobileNetv2', 'MobileNetv3', 'ResNet', 'SENet',
+    'ShuffleNet', 'ShuffleNetv2', 'SwinTransformer', 'VGG', 'ViT', 'ZFNet'
+]
+
+
+def run_training(model_name, train_type, script_dir):
+    """运行指定模型的训练"""
+    model_code_dir = os.path.join(script_dir, model_name, 'code')
+    train_script = os.path.join(model_code_dir, 'train.py')
+    dataset_path = os.path.join(script_dir, 'AlexNet', 'dataset')
+    
+    if not os.path.exists(train_script):
+        print(f"警告: {train_script} 不存在，跳过")
+        return
+    
+    # 切换到模型的code目录
+    os.chdir(model_code_dir)
+    
+    cmd = [
+        'python', 'train.py',  # 使用相对路径
+        '--train-type', train_type,
+        '--dataset-path', dataset_path,  # 保持dataset_path为绝对路径
+        '--gpu', '1'
+    ]
+    
+    print(f"\n开始训练 {model_name} (train_type={train_type})")
+    print(f"工作目录: {os.getcwd()}")
+    print(f"执行命令: {' '.join(cmd)}")
+    
+    try:
+        subprocess.run(cmd, check=True)
+        print(f"{model_name} (train_type={train_type}) 训练完成")
+    except subprocess.CalledProcessError as e:
+        print(f"错误: {model_name} (train_type={train_type}) 训练失败")
+        print(f"错误信息: {str(e)}")
+
+def main():
+    # 获取脚本的绝对路径
+    script_dir = os.path.dirname(os.path.abspath(__file__))
+    original_dir = os.getcwd()  # 保存原始工作目录
+    
+    try:     
+        # 遍历所有模型和训练类型
+        for model in models:
+            for train_type in ['0', '1', '2']:
+                run_training(model, train_type, script_dir)
+    finally:
+        # 恢复原始工作目录
+        os.chdir(original_dir)
+
+if __name__ == '__main__':
+    main()

Image/utils/parse_args.py

@@ -15,4 +15,5 @@ def parse_args():
     parser.add_argument('--poison-ratio', type=float, default=0.1, help='恶意样本比例')
     parser.add_argument('--target-label', type=int, default=0, help='目标类别') 
     parser.add_argument('--train-type',type=str,choices=['0','1','2'],default='0',help='训练类型:0 for normal train, 1 for data aug train,2 for back door train')
+    parser.add_argument('--dataset-path', type=str, default=None, help='本地数据集路径，如果不指定则自动下载')
     return parser.parse_args()

Image/utils/train_utils.py

@@ -260,23 +260,25 @@ def train_model(model, trainloader, testloader, epochs=200, lr=0.1, device='cuda
         logger.info(f'Epoch: {epoch+1} | Test Loss: {test_loss/(batch_idx+1):.3f} | '
                    f'Test Acc: {acc:.2f}%')
         
-        # 创建epoch保存目录
-        epoch_dir = os.path.join(save_dir, f'epoch_{epoch+1}')
-        if not os.path.exists(epoch_dir):
-            os.makedirs(epoch_dir)
+        # 每5个epoch保存一次
+        if (epoch + 1) % 5 == 0:
+            # 创建epoch保存目录
+            epoch_dir = os.path.join(save_dir, f'epoch_{epoch+1}')
+            if not os.path.exists(epoch_dir):
+                os.makedirs(epoch_dir)
+                
+            # 保存模型权重
+            model_path = os.path.join(epoch_dir, 'subject_model.pth')
+            torch.save(model.state_dict(), model_path)
             
-        # 保存模型权重
-        model_path = os.path.join(epoch_dir, 'subject_model.pth')
-        torch.save(model.state_dict(), model_path)
-        
-        # 收集并保存嵌入向量
-        embeddings, indices = collect_embeddings(model, trainloader, device)
-        # 保存嵌入向量
-        np.save(os.path.join(epoch_dir, 'train_data.npy'), embeddings)
-        
-        # 保存索引信息 - 仅保存数据点的索引列表
-        with open(os.path.join(epoch_dir, 'index.json'), 'w') as f:
-            json.dump(indices, f)
+            # 收集并保存嵌入向量
+            embeddings, indices = collect_embeddings(model, trainloader, device)
+            # 保存嵌入向量
+            np.save(os.path.join(epoch_dir, 'train_data.npy'), embeddings)
+            
+            # 保存索引信息 - 仅保存数据点的索引列表
+            with open(os.path.join(epoch_dir, 'index.json'), 'w') as f:
+                json.dump(indices, f)
         
         # 如果是最佳精度，额外保存一份
         if acc > best_acc:
@@ -297,8 +299,8 @@ def train_model(model, trainloader, testloader, epochs=200, lr=0.1, device='cuda
     logger.info(f'最佳测试精度: {best_acc:.2f}%')
     
 def train_model_data_augmentation(model, epochs=200, lr=0.1, device='cuda:0', 
-                                save_dir='./checkpoints', model_name='model_augmented',
-                                batch_size=128, num_workers=2):
+                               save_dir='../model', model_name='model',
+                               batch_size=128, num_workers=2, local_dataset_path=None):
     """使用数据增强训练模型
     
     数据增强方案说明：
@@ -318,6 +320,7 @@ def train_model_data_augmentation(model, epochs=200, lr=0.1, device='cuda:0',
         model_name: 模型名称
         batch_size: 批次大小
         num_workers: 数据加载的工作进程数
+        local_dataset_path: 本地数据集路径
     """
     import torchvision.transforms as transforms
     from .dataset_utils import get_cifar10_dataloaders
@@ -340,7 +343,7 @@ def train_model_data_augmentation(model, epochs=200, lr=0.1, device='cuda:0',
     ])
     
     # 获取数据加载器
-    trainloader, testloader = get_cifar10_dataloaders(batch_size, num_workers)
+    trainloader, testloader = get_cifar10_dataloaders(batch_size, num_workers, local_dataset_path)
     
     # 使用增强的训练数据
     trainset = trainloader.dataset
@@ -352,9 +355,9 @@ def train_model_data_augmentation(model, epochs=200, lr=0.1, device='cuda:0',
     train_model(model, trainloader, testloader, epochs, lr, device, save_dir, model_name,save_type='1')
 
 def train_model_backdoor(model, poison_ratio=0.1, target_label=0, epochs=200, lr=0.1,
-                        device='cuda:0', save_dir='./checkpoints', model_name='model_backdoor',
-                        batch_size=128, num_workers=2):
-    """使用后门攻击训练模型
+                      device='cuda:0', save_dir='../model', model_name='model',
+                      batch_size=128, num_workers=2, local_dataset_path=None):
+    """训练带后门的模型
     
     后门攻击方案说明：
     1. 标签翻转攻击：将选定比例的样本标签修改为目标标签
@@ -374,13 +377,14 @@ def train_model_backdoor(model, poison_ratio=0.1, target_label=0, epochs=200, lr
         model_name: 模型名称
         batch_size: 批次大小
         num_workers: 数据加载的工作进程数
+        local_dataset_path: 本地数据集路径
     """
     from .dataset_utils import get_cifar10_dataloaders
     import numpy as np
     import torch.nn.functional as F
     
     # 获取原始数据加载器
-    trainloader, testloader = get_cifar10_dataloaders(batch_size, num_workers)
+    trainloader, testloader = get_cifar10_dataloaders(batch_size, num_workers, local_dataset_path)
     
     # 修改部分训练数据的标签和添加触发器
     trainset = trainloader.dataset

README.md

@@ -3,14 +3,64 @@ license: mit
 ---
 # 模型训练过程汇总（持续更新中）
 
-本仓库按照模型类别和名称进行组织，具体结构如下：
+本仓库采用扁平化的目录结构和标签系统来组织模型，具体说明如下：
+
+## 仓库结构
+
+- **一级目录**：直接以模型名称命名，例如 `Clone-detection-BigCloneBench`、`GraphMAE_QM9` 等
+- **模型目录结构**：每个模型目录下包含：
+  - `code/`：存放模型相关代码和训练脚本
+  - `model/`：存放模型训练过程和权重文件
+    - 按数据集分类
+    - 训练变体（0：标准训练，1：数据增强，2：后门攻击）
+    - 每个epoch的权重文件(.pth)和embedding(.npy)
+  - `dataset/`：训练数据集（解压或压缩包形式）
+
+## 标签系统
+
+每个模型都具有以下标签属性：
+
+1. **数据类型** (data_type)
+   - 代码 (code)
+   - 文本 (text)
+   - 图像 (image)
+   - 图结构 (graph)
+
+2. **任务类型** (task_type)
+   - 分类 (classification)
+   - 生成 (generation)
+   - 检索 (retrieval)
+   - 相似度计算 (similarity)
+   - 表示学习 (representation_learning)
+   - 自动编码 (autoencoder)
+   - 代码补全 (completion)
+   - 预训练 (pretraining)
+
+3. **领域** (domain)
+   - 代码克隆检测 (code_clone_detection)
+   - 代码搜索 (code_search)
+   - 分子性质预测 (molecular_property)
+   - 代码缺陷检测 (code_defect_detection)
+   - 计算机视觉 (computer_vision)
+   - 移动端计算 (mobile_computing)
+   - Transformer架构 (transformer)
+
+4. **输入/输出类型** (input_type/output_type)
+   - 代码 (code)
+   - 代码对 (code_pair)
+   - 代码token序列 (code_tokens)
+   - 代码排序 (code_ranking)
+   - 自然语言 (natural_language)
+   - 图结构 (graph)
+   - 图像 (image)
+   - 二元标签 (binary)
+   - 类别标签 (class_label)
+   - 分子特征 (molecular_features)
+
+所有模型的元数据和标签信息都存储在 `models.json` 文件中
+
+可以通过运行 `python model_filter.py` 命令来通过标签进行快速检索和筛选。
 
-- **一级目录**：代表不同的模型类别。
-- **二级目录**：在每个模型类别下，进一步细分为具体的模型名称。
-- **三级目录**：在每个模型名称下，包含以下三个部分：
-  - `code`：存放与模型相关的代码和训练脚本。(现增加了训练过程的记录，三种`.log`文件记录训练过程)
-  - `model`：收集的模型训练过程，(一级子目录：对应不同数据集)，(二级子目录：增加了训练变体的记录,子目录0存储正常训练的过程，字目录1存储数据增强训练的过程，字目录2存储后门攻击训练的过程)，(三级子目录：包括每个epoch的`.pth`模型权重文件、`.npy`训练中收集的embedding，以及`index.json`文件，后者包含了embedding对应的数据集中数据点的索引列表)。
-  - `dataset`：提供模型训练使用的数据集，可以是解压后的文件夹形式，或者压缩包形式`dataset.zip`(可以包含多个数据集，需要在代码中进行切换)。
 
 下表汇总了所有收集的模型训练过程信息：
 <table>

count.py

@@ -0,0 +1,10 @@
+import numpy as np
+
+# 请将这里的'your_file_path.npy'替换为你的.npy文件的实际路径
+file_path = '/home/ruofei/RRF/hf-mirror/ttvnet/Image/AlexNet/model/0/epoch_5/train_data.npy'
+
+# 读取.npy文件
+data = np.load(file_path)
+
+# 输出数组的维度
+print("数组维度:", data.shape)

model_filter.py

@@ -0,0 +1,168 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+
+import json
+from typing import Dict, List, Set
+
+class ModelFilter:
+    def __init__(self, json_path: str = "models.json"):
+        with open(json_path, 'r', encoding='utf-8') as f:
+            self.data = json.load(f)
+        self.models = self.data['models']
+        
+        # 收集所有可用的标签
+        self.all_tags = {
+            'data_type': set(),
+            'task_type': set(),
+            'domain': set(),
+            'input_type': set(),
+            'output_type': set()
+        }
+        
+        # 标签类型的中文名称
+        self.tag_names = {
+            'data_type': '数据类型',
+            'task_type': '任务类型',
+            'domain': '领域',
+            'input_type': '输入类型',
+            'output_type': '输出类型'
+        }
+        
+        for model in self.models:
+            tags = model['tags']
+            for tag_type in self.all_tags:
+                if tag_type in tags:
+                    if isinstance(tags[tag_type], list):
+                        self.all_tags[tag_type].update(tags[tag_type])
+                    else:
+                        self.all_tags[tag_type].add(tags[tag_type])
+
+    def print_tag_type_options(self):
+        """打印标签类型选项"""
+        print("\n可选择的标签类型：")
+        for i, (tag_type, name) in enumerate(self.tag_names.items(), 1):
+            print(f"{i}. {name} ({tag_type})")
+
+    def print_tag_options(self, tag_type: str):
+        """打印特定标签类型的可用选项"""
+        print(f"\n=== {self.tag_names[tag_type]} ===")
+        for i, tag in enumerate(sorted(self.all_tags[tag_type]), 1):
+            print(f"{i}. {tag}")
+
+    def filter_models(self, filters: Dict[str, Set[str]]) -> List[Dict]:
+        """根据筛选条件过滤模型"""
+        filtered_models = []
+        
+        for model in self.models:
+            match = True
+            for tag_type, filter_values in filters.items():
+                if not filter_values:  # 如果该类型没有设置筛选条件，跳过
+                    continue
+                    
+                model_tags = model['tags'].get(tag_type, [])
+                if isinstance(model_tags, str):
+                    model_tags = [model_tags]
+                
+                # 检查是否有交集
+                if not set(model_tags) & filter_values:
+                    match = False
+                    break
+            
+            if match:
+                filtered_models.append(model)
+                
+        return filtered_models
+
+    def print_models(self, models: List[Dict]):
+        """打印模型信息"""
+        if not models:
+            print("\n没有找到匹配的模型。")
+            return
+            
+        print(f"\n找到 {len(models)} 个匹配的模型：")
+        for i, model in enumerate(models, 1):
+            print(f"\n{i}. {model['name']}")
+            print(f"   描述: {model['description']}")
+            print(f"   数据集: {model['dataset']}")
+            print(f"   标签:")
+            for tag_type, tags in model['tags'].items():
+                if isinstance(tags, list):
+                    print(f"     - {self.tag_names[tag_type]}: {', '.join(tags)}")
+                else:
+                    print(f"     - {self.tag_names[tag_type]}: {tags}")
+
+def get_user_input(prompt: str, valid_options: Set[str]) -> Set[str]:
+    """获取用户输入的标签"""
+    print(f"\n{prompt}")
+    print("请输入标签编号（多个标签用空格分隔，直接回车跳过）：")
+    while True:
+        try:
+            user_input = input().strip()
+            if not user_input:
+                return set()
+            
+            indices = [int(x) - 1 for x in user_input.split()]
+            selected = set()
+            sorted_options = sorted(valid_options)
+            for idx in indices:
+                if 0 <= idx < len(sorted_options):
+                    selected.add(sorted_options[idx])
+                else:
+                    print(f"无效的选项编号: {idx + 1}")
+                    continue
+            return selected
+        except ValueError:
+            print("请输入有效的数字编号。")
+
+def get_tag_type_choice() -> str:
+    """获取用户选择的标签类型"""
+    tag_types = list(ModelFilter().tag_names.keys())
+    while True:
+        try:
+            choice = input("\n请选择标签类型编号（直接回车开始筛选）：").strip()
+            if not choice:
+                return ""
+            
+            idx = int(choice) - 1
+            if 0 <= idx < len(tag_types):
+                return tag_types[idx]
+            else:
+                print("无效的选项编号，请重试。")
+        except ValueError:
+            print("请输入有效的数字编号。")
+
+def main():
+    print("欢迎使用模型筛选工具！")
+    model_filter = ModelFilter()
+    filters = {} 
+    while True:
+        # 显示标签类型选项
+        model_filter.print_tag_type_options()
+        
+        # 获取用户选择的标签类型
+        tag_type = get_tag_type_choice()
+        if not tag_type:
+            break
+        
+        # 显示所选标签类型的可用选项
+        model_filter.print_tag_options(tag_type)
+        
+        # 获取用户选择的标签
+        selected = get_user_input(
+            f"选择{model_filter.tag_names[tag_type]}标签",
+            model_filter.all_tags[tag_type]
+        )
+        
+        if selected:
+            filters[tag_type] = selected
+    
+    # 筛选并显示结果
+    filtered_models = model_filter.filter_models(filters)
+    model_filter.print_models(filtered_models)
+    # 询问是否继续
+    if input("\n是否继续筛选？(y/n): ").lower() == 'y':
+        main()
+    else:
+        print("\n感谢使用！再见！")
+if __name__ == "__main__":
+    main()

models.json

@@ -0,0 +1,368 @@
+{
+  "models": [
+    {
+      "name": "Clone-detection-BigCloneBench",
+      "tags": {
+        "data_type": ["code"],
+        "task_type": ["classification", "similarity"],
+        "domain": ["code_clone_detection"],
+        "input_type": "code_pair",
+        "output_type": "binary"
+      },
+      "original_path": "Code-Code/Clone-detection-BigCloneBench",
+      "description": "基于大规模代码克隆基准数据集的代码克隆检测模型，任务是进行二元分类(0/1),其中1代表语义等价，0代表其他情况。",
+      "dataset": "BigCloneBench数据集",
+      "epoch": "待上传"
+    },
+    {
+      "name": "Clone-detection-POJ-104",
+      "tags": {
+        "data_type": ["code"],
+        "task_type": ["retrieval", "similarity"],
+        "domain": ["code_clone_detection"],
+        "input_type": "code",
+        "output_type": "code_ranking"
+      },
+      "original_path": "Code-Code/Clone-detection-POJ-104",
+      "description": "基于POJ-104数据集的代码克隆检测模型，任务是识别不同编程题目中相似的代码实现，给定一段代码和一组候选代码，任务是返回具有相同语义的Top K个代码",
+      "dataset": "POJ-104编程题目数据集",
+      "epoch": "待上传"
+    },
+    {
+      "name": "CodeCompletion-token",
+      "tags": {
+        "data_type": ["code"],
+        "task_type": ["generation", "completion"],
+        "domain": ["code_completion"],
+        "input_type": "code_tokens",
+        "output_type": "code_tokens"
+      },
+      "original_path": "Code-Code/CodeCompletion-token",
+      "description": "基于token级别的代码自动补全模型",
+      "dataset": "Java代码token序列数据集",
+      "epoch": "待上传"
+    },
+    {
+      "name": "Defect-detection",
+      "tags": {
+        "data_type": ["code"],
+        "task_type": ["classification"],
+        "domain": ["code_defect_detection"],
+        "input_type": "code",
+        "output_type": "binary"
+      },
+      "original_path": "Code-Code/Defect-detection",
+      "description": "代码缺陷检测模型，通过分析代码来识别潜在的缺陷和错误（进行二元分类(0/1)）",
+      "dataset": "包含缺陷标注的C语言代码数据集",
+      "epoch": "待上传"
+    },
+    {
+      "name": "code-refinement",
+      "tags": {
+        "data_type": ["code"],
+        "task_type": ["generation", "optimization"],
+        "domain": ["code_optimization"],
+        "input_type": "code",
+        "output_type": "code"
+      },
+      "original_path": "Code-Code/code-refinement",
+      "description": "代码优化模型",
+      "dataset": "代码优化前后对数据集（C语言）",
+      "epoch": "待上传"
+    },
+    {
+      "name": "code-to-text",
+      "tags": {
+        "data_type": ["code", "text"],
+        "task_type": ["generation", "translation"],
+        "domain": ["code_documentation"],
+        "input_type": "code",
+        "output_type": "text"
+      },
+      "original_path": "Code-Text/code-to-text",
+      "description": "代码到自然语言的转换模型",
+      "dataset": "多语言代码-文本对数据集",
+      "epoch": "待上传"
+    },
+    {
+      "name": "NL-code-search-Adv",
+      "tags": {
+        "data_type": ["code", "text"],
+        "task_type": ["retrieval", "search"],
+        "domain": ["code_search"],
+        "input_type": "text",
+        "output_type": "code"
+      },
+      "original_path": "Text-code/NL-code-search-Adv",
+      "description": "高级自然语言代码搜索模型，通过计算自然语言查询与代码片段之间的相似性来实现代码搜索",
+      "dataset": "自然语言-(python)代码对数据集",
+      "epoch": "待上传"
+    },
+    {
+      "name": "NL-code-search-WebQuery",
+      "tags": {
+        "data_type": ["code", "text"],
+        "task_type": ["retrieval", "search"],
+        "domain": ["code_search"],
+        "input_type": "text",
+        "output_type": "code"
+      },
+      "original_path": "Text-code/NL-code-search-WebQuery",
+      "description": "基于Web查询的代码搜索模型，该模型通过编码器处理代码和自然语言输入，并利用多层感知器（MLP）来计算相似性得分",
+      "dataset": "Web查询-代码对数据集（CodeSearchNet数据集和CoSQA数据集(python)）",
+      "epoch": "待上传"
+    },
+    {
+      "name": "text-to-code",
+      "tags": {
+        "data_type": ["code", "text"],
+        "task_type": ["generation"],
+        "domain": ["code_generation"],
+        "input_type": "text",
+        "output_type": "code"
+      },
+      "original_path": "Text-code/text-to-code",
+      "description": "自然语言到代码的生成模型",
+      "dataset": "文本描述-代码(c语言)对数据集",
+      "epoch": "待上传"
+    },
+    {
+      "name": "GraphMAE_QM9",
+      "tags": {
+        "data_type": ["graph"],
+        "task_type": ["representation_learning", "autoencoder"],
+        "domain": ["molecular_property"],
+        "input_type": "molecular_graph",
+        "output_type": "graph_embedding"
+      },
+      "original_path": "Graph/GraphMAE_QM9",
+      "description": "在QM9数据集上训练的图掩码自编码器,通过对分子图中的原子的坐标以及类型进行预测实现自监督训练",
+      "dataset": "分子属性预测数据集",
+      "epoch": "待上传"
+    },
+    {
+      "name": "AlexNet",
+      "tags": {
+        "data_type": ["image"],
+        "task_type": ["classification"],
+        "domain": ["computer_vision"],
+        "input_type": "image",
+        "output_type": "class_label"
+      },
+      "original_path": "Image/AlexNet",
+      "description": "2012年获得ImageNet冠军的经典模型，首次证明了深度学习在图像识别上的强大能力。",
+      "dataset": "CIFAR-10数据集",
+      "epoch": "待补充"
+    },
+    {
+      "name": "DenseNet",
+      "tags": {
+        "data_type": ["image"],
+        "task_type": ["classification"],
+        "domain": ["computer_vision"],
+        "input_type": "image",
+        "output_type": "class_label"
+      },
+      "original_path": "Image/DenseNet",
+      "description": "每一层都直接与其他所有层相连，像搭积木一样层层堆叠，可以更好地学习图像特征。",
+      "dataset": "CIFAR-10数据集",
+      "epoch": "待补充"
+    },
+    {
+      "name": "EfficientNet",
+      "tags": {
+        "data_type": ["image"],
+        "task_type": ["classification"],
+        "domain": ["computer_vision"],
+        "input_type": "image",
+        "output_type": "class_label"
+      },
+      "original_path": "Image/EfficientNet",
+      "description": "通过平衡网络的深度、宽度和图像分辨率，用更少的计算量达到更好的效果。",
+      "dataset": "CIFAR-10数据集",
+      "epoch": "待补充"
+    },
+    {
+      "name": "GoogLeNet",
+      "tags": {
+        "data_type": ["image"],
+        "task_type": ["classification"],
+        "domain": ["computer_vision"],
+        "input_type": "image",
+        "output_type": "class_label"
+      },
+      "original_path": "Image/GoogLeNet",
+      "description": "谷歌开发的模型，像多个眼睛同时看图片的不同部分，既省资源又准确。",
+      "dataset": "CIFAR-10数据集",
+      "epoch": "待补充"
+    },
+    {
+      "name": "LeNet5",
+      "tags": {
+        "data_type": ["image"],
+        "task_type": ["classification"],
+        "domain": ["computer_vision"],
+        "input_type": "image",
+        "output_type": "class_label"
+      },
+      "original_path": "Image/LeNet5",
+      "description": "深度学习领域的开山之作，虽然简单但奠定了现代CNN的基础架构。",
+      "dataset": "CIFAR-10数据集",
+      "epoch": "待补充"
+    },
+    {
+      "name": "MobileNetv1",
+      "tags": {
+        "data_type": ["image"],
+        "task_type": ["classification"],
+        "domain": ["computer_vision", "mobile_computing"],
+        "input_type": "image",
+        "output_type": "class_label"
+      },
+      "original_path": "Image/MobileNetv1",
+      "description": "专门为手机设计的轻量级模型，用特殊的卷积方式减少计算量。",
+      "dataset": "CIFAR-10数据集",
+      "epoch": "待补充"
+    },
+    {
+      "name": "MobileNetv2",
+      "tags": {
+        "data_type": ["image"],
+        "task_type": ["classification"],
+        "domain": ["computer_vision", "mobile_computing"],
+        "input_type": "image",
+        "output_type": "class_label"
+      },
+      "original_path": "Image/MobileNetv2",
+      "description": "MobileNet的升级版，增加了特征复用机制，性能更好。",
+      "dataset": "CIFAR-10数据集",
+      "epoch": "待补充"
+    },
+    {
+      "name": "MobileNetv3",
+      "tags": {
+        "data_type": ["image"],
+        "task_type": ["classification"],
+        "domain": ["computer_vision", "mobile_computing"],
+        "input_type": "image",
+        "output_type": "class_label"
+      },
+      "original_path": "Image/MobileNetv3",
+      "description": "结合自动搜索技术的新版本，自动找到最适合手机的网络结构。",
+      "dataset": "CIFAR-10数据集",
+      "epoch": "待补充"
+    },
+    {
+      "name": "ResNet",
+      "tags": {
+        "data_type": ["image"],
+        "task_type": ["classification"],
+        "domain": ["computer_vision"],
+        "input_type": "image",
+        "output_type": "class_label"
+      },
+      "original_path": "Image/ResNet",
+      "description": "通过特殊的快捷连接解决深层网络训练难的问题，可以训练超级深的网络。",
+      "dataset": "CIFAR-10数据集",
+      "epoch": "待补充"
+    },
+    {
+      "name": "SENet",
+      "tags": {
+        "data_type": ["image"],
+        "task_type": ["classification"],
+        "domain": ["computer_vision"],
+        "input_type": "image",
+        "output_type": "class_label"
+      },
+      "original_path": "Image/SENet",
+      "description": "为网络添加了注意力机制，让模型能够关注图片中重要的部分。",
+      "dataset": "CIFAR-10数据集",
+      "epoch": "待补充"
+    },
+    {
+      "name": "ShuffleNet",
+      "tags": {
+        "data_type": ["image"],
+        "task_type": ["classification"],
+        "domain": ["computer_vision", "mobile_computing"],
+        "input_type": "image",
+        "output_type": "class_label"
+      },
+      "original_path": "Image/ShuffleNet",
+      "description": "通过巧妙地打乱和分组计算，实现了手机上的高效运行。",
+      "dataset": "CIFAR-10数据集",
+      "epoch": "待补充"
+    },
+    {
+      "name": "ShuffleNetv2",
+      "tags": {
+        "data_type": ["image"],
+        "task_type": ["classification"],
+        "domain": ["computer_vision", "mobile_computing"],
+        "input_type": "image",
+        "output_type": "class_label"
+      },
+      "original_path": "Image/ShuffleNetv2",
+      "description": "在原版基础上优化设计，速度更快，效果更好。",
+      "dataset": "CIFAR-10数据集",
+      "epoch": "待补充"
+    },
+    {
+      "name": "SwinTransformer",
+      "tags": {
+        "data_type": ["image"],
+        "task_type": ["classification"],
+        "domain": ["computer_vision", "transformer"],
+        "input_type": "image",
+        "output_type": "class_label"
+      },
+      "original_path": "Image/SwinTransformer",
+      "description": "把自然语言处理的先进技术用于图像，通过逐步关注图片不同区域来理解图像。",
+      "dataset": "CIFAR-10数据集",
+      "epoch": "待补充"
+    },
+    {
+      "name": "VGG",
+      "tags": {
+        "data_type": ["image"],
+        "task_type": ["classification"],
+        "domain": ["computer_vision"],
+        "input_type": "image",
+        "output_type": "class_label"
+      },
+      "original_path": "Image/VGG",
+      "description": "用统一的小型卷积核堆叠成深层网络，结构简单但效果好。",
+      "dataset": "CIFAR-10数据集",
+      "epoch": "待补充"
+    },
+    {
+      "name": "ViT",
+      "tags": {
+        "data_type": ["image"],
+        "task_type": ["classification"],
+        "domain": ["computer_vision", "transformer"],
+        "input_type": "image",
+        "output_type": "class_label"
+      },
+      "original_path": "Image/ViT",
+      "description": "把图片切成小块后像读文章一样处理，是一种全新的图像处理方式。",
+      "dataset": "CIFAR-10数据集",
+      "epoch": "待补充"
+    },
+    {
+      "name": "ZFNet",
+      "tags": {
+        "data_type": ["image"],
+        "task_type": ["classification"],
+        "domain": ["computer_vision"],
+        "input_type": "image",
+        "output_type": "class_label"
+      },
+      "original_path": "Image/ZFNet",
+      "description": "通过可视化研究改进的AlexNet，帮助人们理解网络是如何看图片的。",
+      "dataset": "CIFAR-10数据集",
+      "epoch": "待补充"
+    }
+  ]
+}