Skip to content
TheChatArchiveThe Chat Archive
gpt-4o-2024-08-06· OpenAIAi2 WildChat-4.8M (AI2)

BatchSize: 1 Loaded data train: <torch.utils.data.dataloader.DataLoader object at…

  1. User

    BatchSize: 1 Loaded data train: <torch.utils.data.dataloader.DataLoader object at 0x0000023A5D1D4B80> Loaded data test: <torch.utils.data.dataloader.DataLoader object at 0x0000023A5D1D4C70> Learning rate: 0.0001 total trainable params 65054464 Training From Scratch !!! Starting Epoch: 0 QQQQQQQQQQ save npy torch.Size([1, 1, 80, 16]) torch.Size([1, 15, 192, 384]) QQQQQQQQQQQQQ Traceback (most recent call last): File "1.py", line 465, in run() File "1.py", line 458, in run train(device, model, train_data_loader,test_data_loader, optimizer, File "1.py", line 271, in train a, v = model(mel, x) File "H:\Anaconda3\envs\EDTalk\lib\site-packages\torch\nn\modules\module.py", line 1194, in _call_impl return forward_call(*input, **kwargs) File "H:\Anaconda3\envs\EDTalk\lib\site-packages\torch\nn\parallel\data_parallel.py", line 169, in forward return self.module(*inputs[0], **kwargs[0]) File "H:\Anaconda3\envs\EDTalk\lib\site-packages\torch\nn\modules\module.py", line 1194, in _call_impl return forward_call(*input, **kwargs) File "D:\2023project\project\0win_os\05speech_Pro\diff2lip\thirdparty\wav2lip384_my2\models\syncnet.py", line 79, in forward face_embedding = self.face_encoder(face_sequences) File "H:\Anaconda3\envs\EDTalk\lib\site-packages\torch\nn\modules\module.py", line 1194, in _call_impl return forward_call(*input, **kwargs) File "H:\Anaconda3\envs\EDTalk\lib\site-packages\torch\nn\modules\container.py", line 204, in forward input = module(input) File "H:\Anaconda3\envs\EDTalk\lib\site-packages\torch\nn\modules\module.py", line 1194, in _call_impl return forward_call(*input, **kwargs) File "D:\2023project\project\0win_os\05speech_Pro\diff2lip\thirdparty\wav2lip384_my2\models\conv.py", line 28, in forward out = self.conv_block(x) File "H:\Anaconda3\envs\EDTalk\lib\site-packages\torch\nn\modules\module.py", line 1194, in _call_impl return forward_call(*input, **kwargs) File "H:\Anaconda3\envs\EDTalk\lib\site-packages\torch\nn\modules\container.py", line 204, in forward input = module(input) File "H:\Anaconda3\envs\EDTalk\lib\site-packages\torch\nn\modules\module.py", line 1194, in _call_impl return forward_call(*input, **kwargs) File "H:\Anaconda3\envs\EDTalk\lib\site-packages\torch\nn\modules\batchnorm.py", line 171, in forward return F.batch_norm( File "H:\Anaconda3\envs\EDTalk\lib\site-packages\torch\nn\functional.py", line 2448, in batch_norm _verify_batch_size(input.size()) File "H:\Anaconda3\envs\EDTalk\lib\site-packages\torch\nn\functional.py", line 2416, in _verify_batch_size raise ValueError("Expected more than 1 value per channel when training, got input size {}".format(size)) ValueError: Expected more than 1 value per channel when training, got input size torch.Size([1, 1024, 1, 1]) 这个错误如何解决

  2. Assistant

    这个错误是由于在进行批归一化(Batch Normalization)时,输入的批大小为1引起的。批归一化在训练模式下需要多个样本来计算均值和方差,因此当输入的批为1时会出现错误。这里有几种解决方案可以考虑:

    1. 增加批大小(Batch Size):

      • 你可以尝试增大训练时的批大小,使其大于1。修改你的数据加载器,确保batch_size参数大于1。
    2. 使用层归一化(Layer Normalization)或实例归一化(Instance Normalization):

      • 若无法增大批大小,可以尝试使用其他类型的归一化技术,比如层归一化或实例归一化,这些方法不依赖于批大小。
    3. 切换到评估模式:

      • 你可以将模型临时切换到评估模式,通过调用model.eval(),然后进行前向传播。这样,批归一化会使用已统计的均值和方差,而不是批次中的统计数据。但要注意,这仅仅是为了处理这类数据的临时解决办法。
    4. 调整模型结构:

      • 如果上述方法都不可行,可能需要改变你的网络架构,以兼容小批次大小。例如,如果是在更深的层中出现[1, 1024, 1, 1]这种情况,可能需要调整卷积或池化的步长和核大小。
    5. 伪造批次:

      • 在训练时,可以复制样本以形成更大的批次,但这并不是一个规范的方法,因为它可能会导致梯度更新出现偏差。

    请根据项目的具体要求和限制选择最合适的方法。如果你的环境允许,修改batch_size通常是最简单和有效的解决方案。