junyanz/pytorch-CycleGAN-and-pix2pix

Query realted if the model can give RGBA image as output?

roshan2024nar opened this issue · 2 comments

Hello,

just wanted to know can i train the model for taking jpg or png(Transperent) as input and then get result in png(Transperent).....
how should i modify the code ?

Thank you for the help :)

You can train png image if you change --input_nc and --output_nc into 4.
And there might be an converting code in loading the data in file "aligned_dataset.py".
You have to delete the convert code of PIL library.
I hope this help.

Oh, and you might also have to change the normalizing part in file "base_dataset.py"
I remember that the code is for 3 dimension image so you have to add one more dimension when it nomalize it.