Wav2Lip
Wav2Lip copied to clipboard
Rudrabha
训练卡住
edmodel --syncnet_checkpoint_path ./checkpoints/lipsync_expert.pth use_cuda: True total trainable params 36298035 Load checkpoint from: ./checkpoints/lipsync_expert.pth Starting Epoch: 0 use_cuda: True use_cuda: True use_cuda: True use_cuda: True use_cuda: True use_cuda: True use_cuda: True use_cuda: True use_cuda: True use_cuda: True
Solve it？
yes！You need to go into this file and manually set num_workers to 0
PATH：torch\utils\data\dataloader.py
The data I trained was not very good, the cpu was full during training, but the gpu did not move much!
how to set
@Canyao45 Modify to your own virtual path！
PATH：C:\Users\admin\.conda\envs\Wav2LipTrain\Lib\site-packages\torch\utils\data
if persistent_workers and num_workers == 0:
raise ValueError('persistent_workers option needs num_workers > 0')
他好像本来就设置了这样还需要去修改嘛
if persistent_workers and num_workers == 0: raise ValueError('persistent_workers option needs num_workers > 0')他好像本来就设置了这样还需要去修改嘛
你跑起来了吗？
how to set
@Canyao45 Modify to your own virtual path！
PATH：C:\Users\admin\.conda\envs\Wav2LipTrain\Lib\site-packages\torch\utils\data
修改了这个也没有解决问题，发现是enumerate(train_data_loader)这个卡住了，有什么办法解决吗？
