search

crowsonkb / k-diffusion

Karras et al. (2022) diffusion models for PyTorch
MIT License
2.27k stars 373 forks source link

White line augmentation artifact #2

Closed samedii closed 2 years ago

samedii commented 2 years ago

Thank you for open sourcing this! I tried out your implementation of the non-leaky augmentations. In case it's helpful to you I noticed that there seem to be some artifacts created in the augmentation pipeline that will probably not help in training

Left is normal. Right has an added white line image

(I have a different implementation that I built a couple of weeks ago but I didn't do the non-leaky augmentations then. For what it's worth I can say that I've also gotten better results with these techniques than for example v-diff on small real world datasets)

samedii commented 2 years ago

I looked at it and this is just caused by order=3, clip=False and then me converting to uint8 which makes it look white but it's actually overflowed (negative). I don't think this is an issue for you

crowsonkb commented 2 years ago

Ah yes, it is because I don't clamp because I feed it into the training pipeline as floating point. :)

On Sun, Jun 26, 2022 at 3:46 AM Richard Löwenström @.***> wrote:

I looked at it and this is just caused by order=3, clip=False and then me converting to uint8 which makes it look white but it's actually overflowed (negative). I don't think this is an issue for you

— Reply to this email directly, view it on GitHub https://github.com/crowsonkb/k-diffusion/issues/2#issuecomment-1166490175, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABDQ67WW72IMPKUQSNQFLZ3VRAYHNANCNFSM5Z3XF3JA . You are receiving this because you are subscribed to this thread.Message ID: @.***>

samedii commented 2 years ago

Only seen degradations with the non-leaky augmentations so far but I saw that you've made some updates to it so maybe I'll try again