PyTorch DataLoader迭代数轮后抛出RuntimeError报错求助

自定义Dataset代码 #

class Dataset(Dataset):

    'Characterizes a dataset for PyTorch'
    def __init__(self, input_feature_paths, target_feature_folder) -> None:
        self.input_feature_paths = input_feature_paths
        self.target_feature_folder = target_feature_folder
    
    def __len__(self):
        #return sum(1 for _ in self.input_feature_paths)
        return len(self.input_feature_paths)

    def __getitem__(self, index) -> None:
        input_feature_path = self.input_feature_paths[index]
        input_feature = load(input_feature_path, map_location='cpu')
        target_feature_path = self.target_feature_folder / input_feature_path.parts[-1]
        target_feature = load(target_feature_path, map_location='cpu')
        return input_feature.to(dtype=torch.float64), target_feature.to(dtype=torch.float64)

错误栈信息 #

Traceback (most recent call last):
  File "student_audio_feature_extractor.py", line 178, in <module>
    train(dt, input_frame)
  File "student_audio_feature_extractor.py", line 164, in train
    model, train_loss = train_step(model, train_loader, optimizer, criterion)
  File "student_audio_feature_extractor.py", line 80, in train_step
    for input_feature, target_feature in train_loader:
  File "/home/visge/miniconda3/envs/zk_torch/lib/python3.8/site-packages/torch/utils/data/dataloader.py", line 628, in __next__
    data = self._next_data()
  File "/home/visge/miniconda3/envs/zk_torch/lib/python3.8/site-packages/torch/utils/data/dataloader.py", line 1313, in _next_data
    return self._process_data(data)
  File "/home/visge/miniconda3/envs/zk_torch/lib/python3.8/site-packages/torch/utils/data/dataloader.py", line 1359, in _process_data
    data.reraise()
  File "/home/visge/miniconda3/envs/zk_torch/lib/python3.8/site-packages/torch/_utils.py", line 543, in reraise
    raise exception
RuntimeError: Caught RuntimeError in DataLoader worker process 4.
Original Traceback (most recent call last):
  File "/home/visge/miniconda3/envs/zk_torch/lib/python3.8/site-packages/torch/utils/data/_utils/worker.py", line 302, in _worker_loop
    data = fetcher.fetch(index)
  File "/home/visge/miniconda3/envs/zk_torch/lib/python3.8/site-packages/torch/utils/data/_utils/fetch.py", line 61, in fetch
    return self.collate_fn(data)
  File "/home/visge/miniconda3/envs/zk_torch/lib/python3.8/site-packages/torch/utils/data/_utils/collate.py", line 265, in default_collate
    return collate(batch, collate_fn_map=default_collate_fn_map)
  File "/home/visge/miniconda3/envs/zk_torch/lib/python3.8/site-packages/torch/utils/data/_utils/collate.py", line 143, in collate
    return [collate(samples, collate_fn_map=collate_fn_map) for samples in transposed]  # Backwards compatibility.
  File "/home/visge/miniconda3/envs/zk_torch/lib/python3.8/site-packages/torch/utils/data/_utils/collate.py", line 143, in <listcomp>
    return [collate(samples, collate_fn_map=collate_fn_map) for samples in transposed]  # Backwards compatibility.
  File "/home/visge/miniconda3/envs/zk_torch/lib/python3.8/site-packages/torch/utils/data/_utils/collate.py", line 120, in collate
    return collate_fn_map[elem_type](batch, collate_fn_map=collate_fn_map)
  File "/home/visge/miniconda3/envs/zk_torch/lib/python3.8/site-packages/torch/utils/data/_utils/collate.py", line 162, in collate_tensor_fn
    out = elem.new(storage).resize_(len(batch), *list(elem.size()))
RuntimeError: Trying to resize storage that is not resizable

问题原因 #

错误源于加载的Tensor使用了不可调整大小的存储，DataLoader默认的collate_fn在尝试将多个样本拼接成批量Tensor时，无法修改这类存储的大小。另外，若不同样本的Tensor形状不一致，也会触发类似问题。

解决办法 #

1. 转换为可调整的Tensor #

在__getitem__中加载Tensor后，显式创建新的可调整Tensor，确保存储支持修改：

def __getitem__(self, index):
    input_feature_path = self.input_feature_paths[index]
    input_feature = load(input_feature_path, map_location='cpu')
    # 用torch.tensor重新创建可调整Tensor
    input_feature = torch.tensor(input_feature, dtype=torch.float64)
    
    target_feature_path = self.target_feature_folder / input_feature_path.parts[-1]
    target_feature = load(target_feature_path, map_location='cpu')
    target_feature = torch.tensor(target_feature, dtype=torch.float64)
    
    return input_feature, target_feature

或者使用.clone()方法复制Tensor，生成可调整的存储：

input_feature = load(input_feature_path, map_location='cpu').clone().to(dtype=torch.float64)
target_feature = load(target_feature_path, map_location='cpu').clone().to(dtype=torch.float64)

2. 确保样本形状一致 #

检查所有加载的input_feature和target_feature形状完全相同，避免拼接时出错：

def __getitem__(self, index):
    # ... 加载代码 ...
    # 替换为你的预期形状
    expected_input_shape = (128, 100)
    expected_target_shape = (64, 100)
    
    if input_feature.shape != expected_input_shape:
        raise ValueError(f"输入特征形状不匹配：{input_feature.shape} vs {expected_input_shape}")
    if target_feature.shape != expected_target_shape:
        raise ValueError(f"目标特征形状不匹配：{target_feature.shape} vs {expected_target_shape}")
    
    # ... 转换类型并返回 ...

3. 自定义collate_fn #

如果默认拼接逻辑不适用，手动实现批量拼接函数：

def custom_collate(batch):
    # 从batch中取出所有输入和目标特征，堆叠成批量Tensor
    inputs = torch.stack([item[0] for item in batch])
    targets = torch.stack([item[1] for item in batch])
    return inputs, targets

# 创建DataLoader时指定自定义collate_fn
train_loader = torch.utils.data.DataLoader(
    dataset,
    batch_size=32,
    shuffle=True,
    collate_fn=custom_collate
)

内容的提问来源于stack exchange，提问作者tealy