Worked perfectly on cross-Identity reenactment only if the single image is from ffhq datasets

[Shotsordsoo]

And failed every images that aren't. It looks like the face parts are generated properly, yet the locations are wrong:

https://user-images.githubusercontent.com/16696685/206865487-c1aede3d-d56a-4c55-8bba-22c7f9a8eac8.mp4

[FeiiYin]

Since our framework is based on StyleGAN, we need to perform GAN inversion method to reconstruct the identity image. StyleGAN is trained on aligned high-resolution images, i.e. FFHQ. Therefore, GAN inversion methods, along with our method required aligned data.

[Shotsordsoo]

Since our framework is based on StyleGAN, we need to perform GAN inversion method to reconstruct the identity image. StyleGAN is trained on aligned high-resolution images, i.e. FFHQ. Therefore, GAN inversion methods, along with our method required aligned data.

Thx, work like a charm

[FeiiYin]

:-)