Experiment result

[Feiyu-Zhang]

I tried to train on my custom datasets from scratch.
However, the generated image is so bad as follow.

[WonwoongCho]

Hi Feiyu-Zhang,
I apologize for the late reply.

I have not observed those kinds of bad-quality images when I experimented GDWCT using CelebA dataset.

You need to check the validity of the dataset.
I recommend training MUNIT and DRIT by using the dataset.
If those methods are properly trained, it means the dataset is valid, to some extent.

Please leave a reply again, if other methods work.
(If you share the dataset, it would be a better discussion.)