Kim Seonghyeon

@yueyang130 You may need to add CUDA header directory path to CPLUS_INCLUDE_PATH. Maybe $CUDA_HOME/include.

@danielkaifeng Please check your CUDA installations, and set PATH environment variables to the path where nvcc resides.

Using `torch.no_grad` will reduce the memory consumption.

Currently this implementation does not supports class conditions.

Currently this implementation does not support label supervision. But it will be not very hard to add it.

Nice work! Definitely it looks like how this implementation should have been.

Generally EMA improves training stability, especially in GANs.

It is common practice, and you can refer to the paper like this: https://arxiv.org/abs/1806.04498

@LiUzHiAn It means weight will be decayed to 50% when model is trained with 10000 images with batch size 32.

I think some of your image files are not a valid jpeg files.