SHI-Labs / Prompt-Free-Diffusion

Prompt-Free Diffusion: Taking "Text" out of Text-to-Image Diffusion Models, arxiv 2023 / CVPR 2024
https://arxiv.org/abs/2305.16223
MIT License
733 stars 36 forks source link

Can't Reproduce the result with SeeCoder-PA #27

Open beotborry opened 5 months ago

beotborry commented 5 months ago

First of all, congratulations on your wonderful work being accepted to CVPR!

I have a question about the result of image variation with SeeCoder-PA.

I tried to reproduce the result of image variation with "assets/examples/ghibli-input.jpg" using pre-trained SeeCoder-PA without any additional edge condtitioning. But I couldn't get the result shown in your paper.

Could you check if the pre-trained weights uploaded in the huggingface work properly? If so, could you let me know the code to reproduce the results in your paper?