search

sjtuplayer / few-shot-diffusion

[ICCV 2023] Phasic Content Fusing Diffusion Model with Directional Distribution Consistency for Few-Shot Model Adaption
50 stars 2 forks source link

Questions abount the pratical performance of the released checkpoints #11

Closed Ferry-Li closed 4 months ago

Ferry-Li commented 4 months ago

Hi! This work is inspiring and impressive! It is so nice of you to release the code for research. However, I encountered some problems when using the released checkpoints. I downloaded them from target-domain diffusion model and tried to run the generate.py with python generate.py --ckpt_path ckpt/target_domain/ffhq_to_cartoon.pth --img_path source_img --save_path generations/cartoon where I put some pngs into the dir source_img from the ffhq dataset. The output image is confusing: 0.png and I hardly find the "cartoon" style. Are there some mistakes when I generate the cartoon-styled images?

sjtuplayer commented 4 months ago

How about other styles like sketch?

Ferry-Li commented 4 months ago

How about other styles like sketch?

The outputs of the three styles (cartoon, sketch, and Vangogh) are similar as displayed above. Speicifically, sketch: sketch.png Vangogh: Vangogh.png

sjtuplayer commented 4 months ago

How about other styles like sketch?

The outputs of the three styles (cartoon, sketch, and Vangogh) are similar as displayed above. Speicifically, sketch: sketch.png Vangogh: Vangogh.png

Maybe you can check the version of denoising_diffusion_pytorch? Is it 0.28.0? We have rechecked the code and checkpoints, there is no problem when we run it.

Ferry-Li commented 4 months ago

Thank you very much! I checked my version and re-install the 0.28.0. It works.

Thanks again for your help and your amazing work!