I came up with a bit related idea yesterday and found this paper! It's interesting.
Didn't you compute FID/KID score for the CycleGAN model initialized by your proposed method in the downstream task such as horse2zebra, following recent works on improving CycleGAN, such as this paper ?
I came up with a bit related idea yesterday and found this paper! It's interesting. Didn't you compute FID/KID score for the CycleGAN model initialized by your proposed method in the downstream task such as horse2zebra, following recent works on improving CycleGAN, such as this paper ?