search

aitorzip / PyTorch-CycleGAN

A clean and readable Pytorch implementation of CycleGAN
https://arxiv.org/abs/1703.10593
GNU General Public License v3.0
1.19k stars 283 forks source link

How did you implement PatchGAN?? #1

Open kenoharada opened 6 years ago

kenoharada commented 6 years ago

Thank you so much for implementing CycleGAN in pytorch more readable!

I would like to know which part is PatchGAN??

olivier-gillet commented 5 years ago

PatchGAN corresponds to the discriminator part :
https://github.com/aitorzip/PyTorch-CycleGAN/blob/67da8f9e2b69bd68763451803c7700aaccc92f18/models.py#L64

Einstellung commented 4 years ago

I don"t think we use PatchGAN, can we think avg_pool2d means PatchGAN?

alexander-soare commented 4 years ago

Hi all, I was stuck here too but I've figured it out. After the last conv layer of the PatchGAN (before average pool) the receptive field size is 70. So each neuron on the single channel feature map (which is 30x30) coming out of that conv layer has information from a 70x70 patch of the input. The corresponding patches overlap one another on the input.