search

BadourAlBahar / pose-with-style

[SIGGRAPH Asia 2021] Pose with Style: Detail-Preserving Pose-Guided Image Synthesis with Conditional StyleGAN
https://pose-with-style.github.io/
MIT License
266 stars 41 forks source link

Silhouettes generation issue #14

Closed harryzhangOG closed 1 year ago

harryzhangOG commented 2 years ago

Hi, Thanks for making the code available. I have a question regarding the silhouettes generation. I followed your steps and managed to run PGN inference on the deep fashion dataset. However, it is very slow to generate the segmasks (~20s per image) and often runs with out-of-memory warnings. For the reference, I am using nvidia V100 16Gb so it should be enough... Thus, would it be possible for you to share the segmasks to us via Google drive at your convenience? If not, would it be possible for you to share some pointers on how you made it run fast enough for all the 40k+ images? Thank you so much.

darioshehni commented 2 years ago

I had this same issue. You have to change the code so it actually uses the GPU. It's been a while so I'm not quite sure how I resolved it.

harryzhangOG commented 2 years ago

Hi @darioshehni. thanks for the reply, could you share some pointers on how u resolved it? thanks.

darioshehni commented 2 years ago

Confirm that the GPU is actually recognised and use .to(device) to push the network and all images to the GPU.

BadourAlBahar commented 1 year ago

Thank you very much for your interest. Here are the silhouettes for the training set and testing set.