Double Discriminators for Each Domain

[unlugi]

First of all, thank you for your work. It really helped me. My question is regarding the double discriminators used in the paper. If I am not wrong, There are a total of 4 discriminators, 2 for domain A where one is for translated images and one for image generation using random noise vectors; it is the same for domain B. But from your implementation, I think you use the same set of discriminators for random noise generated images. https://github.com/taki0112/DRIT-Tensorflow/blob/384f6aac3e91898ee400c57418d2e7b3d6df1916/DRIT.py#L432-L434

Also, in the original implementation, the batch of images is halved; the first half is used for translating images and the second half is used for generation from noise. I think you have omitted that. Can you explain if I am wrong?

Thank you so much again.