search

facebookresearch / pytorch_GAN_zoo

A mix of GAN implementations including progressive growing
BSD 3-Clause "New" or "Revised" License
1.61k stars 271 forks source link

where does the input dimension changes in discriminator? #134

Closed Yoonho-Na closed 2 years ago

Yoonho-Na commented 3 years ago

I'm using PGGAN and I found out that in the optimizeParameters function on base_GAN.py, the self.real_input shape changes right after it enters discriminator(before going through any layers). the self.real_input had shape of [16, 3, 4, 4] and I printed out the input x, in the discriminator forward function It printed [4,3,4,4] . why is that and where does it change the dimension?

LiUzHiAn commented 2 years ago

Did you use multi-GPUs to train? Maybe 4 GPUs I guess?

Yoonho-Na commented 2 years ago

Did you use multi-GPUs to train? Maybe 4 GPUs I guess?

Yes that was the problem.