Open piwawa opened 2 months ago
DataSet里的continue打一下断点,原项目写的数据加载方式总会跳过异常,变成无效死循环
DataSet里的continue打一下断点,原项目写的数据加载方式总会跳过异常,变成无效死循环
Syncnet能训了,不过loss一直在0.69。
python hq_wav2lip_sam_train.py
现在训wav2lip又出问题了,用的一模一样的数据集,一直卡在这不动,4090显卡,200多G内存,卡好几天没反应。
The epoch 0 has been going on for 2 days. I have filled in the path correctly in
train.txt
andtext.txt
, the dataset has 330k videos and has been preprocessed. If I select some data fromtrain.txt
, it can start training immediately. Why can't use a large dataset to train?