search

ermongroup / SDEdit

PyTorch implementation for SDEdit: Image Synthesis and Editing with Stochastic Differential Equations
https://sde-image-editing.github.io/
MIT License
966 stars 90 forks source link

Wrong repository for training the models? #8

Closed alvarogonjim closed 2 years ago

alvarogonjim commented 3 years ago

HI @chenlin9

First of all, congratulations on such a fantastic project. I had a doubt regarding the training models. As I understood, you employed a trained SDE model to perturb the target image into a noise image, and after that, you reverse the stochastic process to get the final image. In that case, we should use the repository of @yang-song https://github.com/yang-song/score_sde (this one for PyTorch implementation) to train SDE models instead of https://github.com/ermongroup/ddim right?

Thank you so much.

xiaoiker commented 2 years ago

Seems @alvarogonjim is right. Also confuse about this.

KellyYutongHe commented 2 years ago

Hi @alvarogonjim and @xiaoiker! There are two versions of SDE/diffusion models that are compatible with our method: VP (ddpm, ddim, etc) and VE (score SDE from @yang-song ). You can use either as the pretrained model and then use the corresponding algorithm to reverse the image. The algorithm for VE is described in Algorithm 1 and 2 and the algorithm for VP is in the appendix in Algorithm 3 and 4. Feel free to check both out!