search

danielroich / PTI

Official Implementation for "Pivotal Tuning for Latent-based editing of Real Images" (ACM TOG 2022) https://arxiv.org/abs/2106.05744
MIT License
897 stars 112 forks source link

inversion on whole body images #36

Closed onyekaokonji closed 2 years ago

onyekaokonji commented 2 years ago

I'm trying to perform inversion on whole body images as opposed to faces. Looking at the inference notebook you shared, I'm guessing the preprocessing function which receives input from align_faces will need a new function called align_body (for example) to provide input for a body image.

Or will the best solution be to skip the preprocessing step all together?

danielroich commented 2 years ago

I think you will need an aligning function as you said

the main reason is because StyleGAN 2 is designed for aligned images.

Maybe you can try StyleGAN 3 if you are not interested in developing such a function. see https://github.com/yuval-alaluf/stylegan3-editing for more details. They have added PTI inversion as well.

I think there are papers that use StyleGAN and full-body, maybe they will help guide you

best of luck!

onyekaokonji commented 2 years ago

Thank you! I eventually found StyleGAN-Human's repo in which they designed a function for PTI inversion of body images. Of course using an alignment function for bodies.

sawk1 commented 1 year ago

Thank you! I eventually found StyleGAN-Human's repo in which they designed a function for PTI inversion of body images. Of course using an alignment function for bodies.

Hello, im trying to find it, but bad luck so far(

hajiarab1988 commented 11 months ago

Thank you! I eventually found StyleGAN-Human's repo in which they designed a function for PTI inversion of body images. Of course using an alignment function for bodies.

Hello, im trying to find it, but bad luck so far(

https://github.com/stylegan-human