pfnet-research/sngan_projection

Use spectral normalization in generator for image enhancement

ligua opened this issue · 0 comments

ligua commented

Hi, thank you for spending time reading this post.

I am implementing spectral normalization for my image enhancement project. I have read your paper and a paper titled "self-attention GAN" which suggest using SN in the generator(which ensures the Lipschitz continuity). However, when I am sythesizing the image. The result seems great.

image

However, when I tried to run the model on the test image, the model produces greenish image with clear texture. I am wondering what may be cause of this.