DBMHT: A double-branch multi-hypothesis transformer for 3D human pose estimation in video

Xuezhi Xiang, Xiaoheng Li, Weijie Bao, Yulong Qiao, Abdulmotaleb El-Saddik

Abstract: Highlights•We propose a Dual-Branch Transformer structure to capture both spatial and temporal information.•We propose a lightweight Multi-Hypothesis Merging module to improve the performance of 3D pose estimation.•We propose DBMHT which achieves competitive performance on Human3.6M and MPI-INF-3DHP dataset.