doc for check images script

This commit is contained in:
Victor Hall 2024-05-05 20:52:54 -04:00
parent ad4ade7781
commit d29368eb85
1 changed files with 9 additions and 1 deletions

View File

@ -11,4 +11,12 @@ See [VRAM](VRAM.md) for more info.
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.
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.