search

advimman / lama

🦙 LaMa Image Inpainting, Resolution-robust Large Mask Inpainting with Fourier Convolutions, WACV 2022
https://advimman.github.io/lama-project/
Apache License 2.0
8.06k stars 855 forks source link

Discriminator config #57

Closed lixiaocong-svg closed 2 years ago

lixiaocong-svg commented 2 years ago

Hello, your paper mentions that the discriminator uses Fourier or dilated convolution, but I see that the discriminator is set to'pix2pixhd_nlayer' in the config files. What is the reason?

windj007 commented 2 years ago

Hi! No, we do not use FFC/dilations in the discriminator - and the paper should not contain such info. The configs are correct.

Could you please point us to the part of the paper which causes such a confusion?

lixiaocong-svg commented 2 years ago

'TheφHRF(x)can be implemented using Fourier or Dilated convolutions.' Sorry, I may have misunderstood.The cause of my misunderstanding is HRFPL loss. But there is FFCNLayerDiscriminator in the ffc file, but it seems that it cannot be called directly. Can you tell me the reason? Finally, thank you very much for your reply

windj007 commented 2 years ago

FFCNLayerDiscriminator

We just forgot to remove that piece of code before code release. It was not used to train any final models. We tried it but it was not very helpful.