🤖 FaceAnything: 4D Face Reconstruction from Video
The FaceAnything model has been released for 4D face reconstruction from any sequence of images or video. The method predicts depth and canonical coordinates, ensuring stable 3D geometry without the use of specialized rigs.
🌍 A new feed-forward approach combines reconstruction and tracking into a single task, providing high temporal consistency without heavy optimization cycles.
👤 It enables the creation of detailed 3D models and digital avatars using ordinary phone videos, simplifying pipelines for CGI and AR.
Source 1: https://github.com/kocasariumut/FaceAnything Source 2: https://huggingface.co/UmutKocasari/FaceAnything