Open modpotato opened 6 months ago
I also have a side issue sort of related. When trying to train 113,246 images, if one single image is bad it messes up the latent caching, and have to delete it and start over. This isn't ideal and should really find a way to bypass corrupted images automatically.
could not load image, it might be corrupted: H:/Dataset\145439\2966342.jpg | 20652/113246 [44:55<3:02:18, 8.47it/s] caching: 18%|███████████▋ | 20652/113246 [44:55<3:21:23, 7.66it/s] epoch: 0%| | 0/180 [45:03<?, ?it/s] Traceback (most recent call last): File "C:\OneTrainer\modules\ui\TrainUI.py", line 478, in __training_thread_function trainer.train()
What happened?
attempted to open dataset tools to automatically tag all images loaded, tkinter fails to create enough menus for them on the left where you can change between them.
What did you expect would happen?
menus would load properly and handle tagging at scale
Relevant log output
Output of
pip freeze
distlib==0.3.7 filelock==3.13.1 platformdirs==4.1.0 pypresence==4.3.0 virtualenv==20.25.0