Skip to content

Latest commit

 

History

History
22 lines (12 loc) · 1.17 KB

TROUBLESHOOTING.md

File metadata and controls

22 lines (12 loc) · 1.17 KB
Raw

UserWarning: None of the inputs have requires_grad=True. Gradients will be None

Not an error, this happens during the first time a set of samples are generated. Ignore.

CUDA out of memory

See VRAM for more info.

** Some images are smaller than the target size, consider using larger images

** Check logs\project_abc_sd21_20230301-122543\undersized_images.txt for more information.

Check the file linked, this is a warning that some of the imags are smaller than the resolution you're using to train.

The best option is to source the originals again at a higher size. Or you can ignore it, use a high quality upscaler, remove the images, or reduce the resolution you're training. You might be ok if it is very small percentage of your dataset, but it is something you should check on.

Errors with PIL

Run this script to check images if your training crashes and PIL or you get any errors that seems image related:

python scripts/check_images.py --data_root "C:\my_training_data"

This will help identify invalid or malformatted images. You can then try using a typical image editor (Photoshop, Gimp, Paint.net, etc.) to fix them, or simply delete them.