Jun-Yan Zhu

probably multiple distortions.

Yes. In your case, the classification task for D might be too easy. I will try to reduce the learning rate of D.

Hard to say. The extra objects in X (e.g., architecture) might disappear after translation.

Not sure. You may want to try our Python [code](https://github.com/junyanz/pytorch-CycleGAN-and-pix2pix) on Windows. It will still take efforts to run it on Windows.

I guess you failed to create the output directory. See more details here. Make sure you can create this text file under the directory. ``` fd = io.open(paths.concat(opt.checkpoints_dir, opt.name, 'opt_'...

@qianxiao111 It seems that you haven't installed Torch. Also, feel free to try out Python [version](https://github.com/junyanz/pytorch-CycleGAN-and-pix2pix).

@xixilzh It seems that the program fails to create the directory and texture file. Please double check if you have the permission. ```python fd = io.open(paths.concat(opt.checkpoints_dir, opt.name, 'opt_' .. mode...

Cool results. A few comments: - It seems that geometric consistency loss used in Muller et al. might be a good solution to the deformity issue. Does it help your...

Looks good. You might want to use instancenorm.

That is one possibility. You can also manually add some preprocessing code before you call the netD.forward. Like [here](https://github.com/junyanz/CycleGAN/blob/master/models/cycle_gan_model.lua#L141).