Jun-Yan Zhu

Great work. Thanks for sharing your results with us.

It is not available. You can adopt other training scripts such as this [one](https://github.com/junyanz/BicycleGAN/blob/master/scripts/train_facades.sh).

It's strange. Have you set CUDA_VISIBLE_DEVICES?

I just downloaded the code and it works for me. Could you check out the latest code?

@pathak22 I think it should be pow(2) according to Equation (10) in the VAE [paper](https://arxiv.org/pdf/1312.6114.pdf).

Your D_X_loss might be too small. It indicates that the discriminator is winning easily. The loss in GANs training might not be as informative as classification tasks. I recommend that...

Sorry that we haven't worked with Lua/Torch for years and cannot help you debug it. Please use the PyTorch [version](https://github.com/junyanz/pytorch-CycleGAN-and-pix2pix), which is better supported.

It can be applied to SR and inverse halftone. But it might not work as well as supervised methods (such as supervised SR). For SR, it's quite easy to get...

Nothing special. We followed the architecture in [DCGAN](https://github.com/pytorch/examples/blob/master/dcgan/main.py). Probably better discriminator designs now.

Sounds like a good idea. Will have a look when we have some free time. @taesungp