search

zhujiapeng / LowRankGAN

[NeurIPS 2021] Low-Rank Subspaces in GANs
https://zhujiapeng.github.io/LowRankGAN/
MIT License
124 stars 5 forks source link

The problem of your latent space? #7

Open Kihensarn opened 2 years ago

Kihensarn commented 2 years ago

After reading your paper and code, I think you use the W space. Is that true? Can I use this method to find an editing direction in W+ space?

zhujiapeng commented 2 years ago

Actually, I did not do such experiments. However, there are two ways you can try. One is to find the directions in the W space and then directly edit the images in the W+ space. The other one is to find the directions in the W+ space (e.g., computing the Jacobians in the W+ space) and then edit the images in the W+ space.

Kihensarn commented 2 years ago

Thanks for your reply. But how can I find the directions in W+ space? Should I modify the code?

zhujiapeng commented 2 years ago

Yes, change the W space to W+ space when computing Jacobians.