search

tencent-ailab / IP-Adapter

The image prompt adapter is designed to enable a pretrained text-to-image diffusion model to generate images with image prompt.
Apache License 2.0
5.08k stars 331 forks source link

Face segmentation #304

Open zechenghe opened 7 months ago

zechenghe commented 7 months ago

Which segmentation model was used for training IPA-Full-Face? I saw a comment in ip_adapter-full-face_demo.ipynb "You should use a cropped face image, segment the face, and fill the background with pure white (255, 255, 255)", did you also segment the face and fill the background with white during training? Thanks!

xiaohu2015 commented 7 months ago

you can use some Background Removal model, e.g. https://huggingface.co/briaai/RMBG-1.4

zechenghe commented 7 months ago

you can use some Background Removal model, e.g. https://huggingface.co/briaai/RMBG-1.4

Does training also remove the background?

xiaohu2015 commented 7 months ago

yes