Thin-Plate-Spline-Motion-Model

Why we included this project

Face reenactment pipelines often come down to one problem: taking a still portrait and making it follow the motion of a driving video without losing the person's identity. This is the official code release for the CVPR 2022 Thin-Plate Spline Motion Model paper, and it handles that problem cleanly enough to be a useful reference for talking-head and lip-sync work. You get training configs, dataset preparation scripts for VoxCeleb and TaiChi, pretrained weights, and working web and Colab demos, so you can judge the output before wiring it into your own stack. The trade-off is that it is research code, not a turnkey product: expect to handle dataset setup and tuning yourself. That is a fair price for a well-documented implementation of a peer-reviewed method.