Issue with GAN model's wierd output

[wise moon]

I am training a GAN model following CycleGAN architecture. However, I am trying to reuse the down-sampling layers inside the discriminator as an encoder, and share it with the generator.
This is the output from 17k-th iteration, and the dataset I am using is selfie2anime. The issue is that the generator keep output this style without any change .
I don't really know what could be the issue as they are way too different from the target images.
If you can, please explain it to me. I am still very new to GAN.
Thank you.

[vagrant tinsel]

The issue is that the generator keep output this style without any change .
This seems like the classic "mode collapse" problem in training GANs as a result of the lack of diversity during training. This blog post https://pub.towardsai.net/gan-mode-collapse-explanation-fa5f9124ee73 explains it well.
You can also take a look at some problems and how to address them here https://developers.google.com/machine-learning/gan/problems