Progressive Fine-Tuning Stage 1 & 2 about dualstylegan HOT 6 CLOSED

Progressive Fine-Tuning Stage 1 & 2 transfers the color and some structure style of realistic face images.
The extrinsic style here is the style of realistic face images.
Please refer to our paper for the details.

from dualstylegan.

But why the results of Stage 1 & 2 are seemed as the color and structural transferred styles of the example seen in the lower-left corner? I'm a little confused...

from dualstylegan.

Because both the realistic face images and the cartoon images use the same psp encoder.
It means psp encoder will map both the realistic face images and the cartoon images into a shared style space, which captures the color and structural styles of the input image.
Even Stage 1 & 2 uses the extrinsic style code of realistic face images for training, during testing we can still use the extrinsic style code of cartoon images.

from dualstylegan.

Does this mean：

1. psp encoder wasn't trained on cartoon images?
2. Stage 1 & 2 don't show the style of cartoon style during training?

from dualstylegan.

psp encoder is pre-trained on FFHQ
Stage 1 & 2 only uses the realistic face image. No cartoon images are used.

This is an example of the training images

And this is an exmaple of the testing images

from dualstylegan.

I got it! Thanks a lot!!!

from dualstylegan.