[pix2pixHD] grey scale image generated in png image training model

Hi, I'm having trouble with pix2pixHD model while training png images.

I fixed the code a little bit to train properly, such as normalizing part and --no_vgg_loss option.

but when the image is generated, it came out only in grey scale image.

every images in checkpoints folder is in grey scale, including input_label, real_image etc.

I found the other repo someone has the same issue but it's not solved yet.

Is there anyone who can fix this issue?

--- the code that I changed in ./data/aligned_dataset line 42/51/63 ...convert('RGB') -> convert('RGBA')

in ./data/base_dataset.py line 56, 57, 61 ...[transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))] -> [transforms.Normalize((0.5, 0.5, 0.5, 0.5), (0.5, 0.5, 0.5, 0.5))]

trained in option --nc_input 4 --nc_output 4 --no_vgg_loss --no_instance