seitaroshinagawa / chainer-stargan Goto Github PK

Hello.
Your implementation helps me a lot since this is my first time to code in Chainer.
It seems that my implementation works not bad in reconstruction loss's convergence.
Here is the change in my implementation:

I use original BatchNormalization in Chainer instead of InstanceNormalization from crcrpar on GitHub, since I find his IN results in the failure of L_rec's convergence. Although BN's result is not so good compared to official version somehow, at least it works I think.
(Although I only test on my own dataset..)

Hope this can be helpful to you !

Hi,

Thank you for your nice work.
Would you please release your pre-trained models?

Cheers!