search

lucidrains / denoising-diffusion-pytorch

Implementation of Denoising Diffusion Probabilistic Model in Pytorch
MIT License
8.05k stars 1.01k forks source link

Why self-conditioning? #94

Open Silverster98 opened 2 years ago

Silverster98 commented 2 years ago

Are there some works or explanations demonstrating the effectiveness of self-conditioning implementation? Or it is just an empirical trick?

https://github.com/lucidrains/denoising-diffusion-pytorch/blob/6e8a0f208265a6e58a3e3d500fdc9f41a6060080/denoising_diffusion_pytorch/denoising_diffusion_pytorch.py#L656-L668

yiyixuxu commented 2 years ago

see paper here https://arxiv.org/abs/2208.04202

tuttyfrutyee commented 1 year ago

This (self-conditioning) does not work for me btw (train loss insists on not decreasing), but the data I work with might be too noisy compared to natural images.