search

Janspiry / Palette-Image-to-Image-Diffusion-Models

Unofficial implementation of Palette: Image-to-Image Diffusion Models by Pytorch
MIT License
1.52k stars 203 forks source link

what's the valid mask #79

Open ooochen-30 opened 1 year ago

ooochen-30 commented 1 year ago

@Janspiry i want to train a inpainting diffuison by my own mask, some of them can get rational result but more of them get a black image. i wonder what's the valid mask, can someone give me a help or have somebody miss the same problem

Liiiii2101 commented 1 year ago

Hey, have you figured out the solution? I'm working on similar tasks

ooochen-30 commented 1 year ago

i solve my problem by changing the picture storage format from JPEG to PNG. As the jpg format does not guarantee the binarization of images.

Liiiii2101 commented 1 year ago

Thanks! will give it a try. Can I ask how's the model performance for you?

ooochen-30 commented 1 year ago

i load the pretrained model of inpainting_places and use 7w+ data, only after serveral epochs it perform great

xmhGit commented 9 months ago

So, I think we need to define a new function in 'mask.py' to load our own masks, right? And I'm wondering if this generative way can maintain the diversity (that possessed by diffusion models) or not. Hoping to get your insights about that.