When you reproduced Stargan, did you use the pytorch version of the code or the tensorflow version of the code? At the same time, should the labels input during training be represented by -0.5 and 0.5 like STGAN? thank you for your reply .