DH-AUG: DH Forward Kinematics Model Driven Augmentation for 3D Human Pose Estimation

Due to the lack of diversity datasets, generalization ability pose estimator is poor. To solve this problem, we propose a augmentation solution via DH forward kinematics model, which call DH-AUG. We observe that previous work all based on single-frame augmentation, if it directly applied video estimator, there will be several previously ignored problems: (i) angle ambiguity in bone rotation (multiple solutions); (ii) generated skeleton lacks movement continuity. these problems, special generator called DH-generator. Extensive experiments demonstrate DH-AUG can greatly increase estimator. In addition, when 3D our method outperforms best method. The source code has been released at https://github.com/hlz0606/DH-AUG-DH-Forward-Kinematics-Model-Driven-Augmentation-for-3D-Human-Pose-Estimation .