3DDFA

Why we included this project

This is the PyTorch reimplementation of a TPAMI 2017 method for face alignment across the full pose range, and it does more than locate landmarks. Given a single photo it predicts dense 3D face vertices and 68 landmarks, and it can render the reconstructed mesh so you can see what the model thinks the face looks like. The code is compact enough to read through, which matters if you want to understand how a 3DMM-based fitting pipeline is put together instead of treating it as a black box. It is a reasonable starting point for AR filters, avatar generation, or head-pose estimation work, though the repo has not seen active updates since 2022, so you will likely need to adapt it to a current PyTorch version.