search

ChenWu98 / cycle-diffusion

[ICCV 2023] A latent space for stochastic diffusion models
Other
575 stars 35 forks source link

Does the unpaired image-to-image translation works with Hugging Face Stable Diffusion checkpoints? #29

Closed nityanandmathur closed 8 months ago

nityanandmathur commented 9 months ago

Hi! I am trying to run unpaired image-to-image translation on a custom dataset.

I am using Hugging Face Unconditional Image Generation Pipeline to train a DDPM model on both domains.

How can I use the saved checkpoints, which are stored as safetensors for the unpaired I2I translation?

ChenWu98 commented 9 months ago

Hi! Sorry but I'm not that familiar with safetensors. We trained the model using this repo, so if you can get your weights somehow loaded with codes in that repo, it should also work for ours.

nityanandmathur commented 9 months ago

Thank you for the information. I'll check it out.

Also, does your method supports unpaired image-to-image translation using LDMs?

ChenWu98 commented 9 months ago

Yeah it does, though it's not reported in the paper. You may refer to config/experiments/translate_ffhq256_to_celeba256_latentdiff_ddim_eta01.cfg for details. I think simply setting RUN_NAME=translate_ffhq256_to_celeba256_latentdiff_ddim_eta01 will probably work.

nityanandmathur commented 9 months ago

Thank you for the information. Could you please let me know which repository did you use to train the LDM?

ChenWu98 commented 9 months ago

I used the pretrained models from https://github.com/CompVis/latent-diffusion.

nityanandmathur commented 8 months ago

Thank you for all the information!