A close look on Paper "Few-shot Knowledge Transfer for Fine-grained Cartoon Face Generation"
Few-shot Knowledge Transfer for Fine-grained Cartoon Face Generation
Basic model
As mentioned above, we first train an image translation model with fX0;Y0g. Some unsupervised methods like CycleGAN [24], UNIT [15] are all suitable.
Here we utilize an open source project for its impressive performance, which is modified from U-GAT-IT[13].
In this project, a face-ID loss are introduced to further improve the performance in response to faces. More details can be found on the project homepage.