search

kkanshul / finegan

FineGAN: Unsupervised Hierarchical Disentanglement for Fine-grained Object Generation and Discovery
http://krsingh.cs.ucdavis.edu/krishna_files/papers/finegan/index.html
BSD 2-Clause "Simplified" License
277 stars 43 forks source link

Training for parent discriminator #11

Closed MikeXuQ closed 4 years ago

MikeXuQ commented 4 years ago

Hi, thanks for publishing your code. But after I reading the code, it makes me confused that the parent discriminator seems not to be updated anymore after initialized. It seems to be different from the paper. Have I ever missed anything important? Looking forward to your answer, thanks!

kidach1 commented 4 years ago

hi, parent discriminator seems to be updated at line 335 of train.py through information loss backward. (and as written in the paper, adversarial loss isn't used in the parent stage.)

utkarshojha commented 4 years ago

Hi MikeXuQ,

The parent discriminator doesn't get trained as a usual discriminator. Instead, it gets optimized for the mutual information loss at the parent stage, along with the generator weights.

MikeXuQ commented 4 years ago

Hi MikeXuQ,

The parent discriminator doesn't get trained as a usual discriminator. Instead, it gets optimized for the mutual information loss at the parent stage, along with the generator weights.

Thanks a lot!