A question about the model you trained

[Zihan-Liu-westlake]

Hi,

I recently compared some adain implementations on github. I found your trained model surprisingly much better than several versions based on pytorch. I think your style transfer model has some kind of image understanding or instance segmentation effect. For example, it can still keep a girl's dress the same color after the style transfer (Instead of going to a colorful dress like a rainbow).

I was trying to embed your code in our model as an augmentation method. Unfortunately, I don't know how to modify the tensorflow code or how to change the inference section in your code to high-throughput (batch-wise I mean). I want to know which dataset you trained the model on, so I can try training it myself with the pytorch version. If there are any special hyperparameters that need attention, please mention them.

Many thanks!