The fake image is not consistent with the original one

[QiYang03101994]

Hello, I am working on the medical image translation. Cycle gan is the useful tool here. But I meet the problem. I used orginal code with modification increasing lambda weight for Cycle loss from 10 to 20. But I notice that generated image is not always consistent with the orignal one (like one part lossing). Is there any way helping me locating the issue ? It is not easy for me to pose image due to the privacy.

[junyanz]

It is hard to tell without knowing your data and task. Feel free to send an email to Taesung and me.