Handling grayscale image

[jyu-theartofml]

For the users who curate their own datasets to train the model, sometimes there will be grayscale images. In that case, I've ran into the error RuntimeError: output with shape [1, 1024, 1024] doesn't match the broadcast shape [3, 1024, 1024] from pro_gan_pytorch-examples/implementation/data_processing/DataLoader.py. I would like to open a PR and make some suggested changes.