The result contains serious noise

VinAIResearch / WaveDiff

Official Pytorch Implementation of the paper: Wavelet Diffusion Models are fast and scalable Image Generators (CVPR'23)

GNU Affero General Public License v3.0

376 stars 29 forks source link

The result contains serious noise #6

Closed LiuYueXian123 closed 1 year ago

LiuYueXian123 commented 1 year ago

I tried to reproduce the effect in the paper, I did not change any parameters, but the result contains serious noise. Does anyone reproduce the effect cleanly, or what can I do to improve it.

hao-pt commented 1 year ago

Could you please give us the details of your test script? What model you are using? (our pretrained models or your reproduced ones)

hao-pt commented 1 year ago

Without modifying any hyper-params, here are the results of our sampled images. As shown, there are no noise artifacts as you encounter. I decide to close this issue as it is inactive for a while. Feel free to reopen in case you need further support. Cheer!

samples_cifar10 samples_celeba_256