DECA

Why we included this project

DECA is the official PyTorch implementation of the SIGGRAPH 2021 paper, and it's a solid reference for monocular 3D face reconstruction: from a single photo it returns a head model with pose, shape, and lighting, along with fine surface detail like wrinkles. The clever part is how it separates a person's stable identity shape from expression-dependent wrinkles, which means you can reanimate the mesh with new poses and expressions and get realistic deformations instead of a static head being repositioned. Inference code, a Colab notebook, and a trained model are included, so it's practical to test on your own photos, including the kind of in-the-wild shots with odd lighting or occlusion that it was designed to handle. One thing to check before you commit: the code ships under a non-commercial research license from Max-Planck, which rules out most commercial work.