A Lightweight Graph Transformer Network for Human Mesh Reconstruction from 2D Human Pose

  • 2021-11-24 18:48:03
  • Ce Zheng, Matias Mendieta, Pu Wang, Aidong Lu, Chen Chen
  • 2

Abstract

Existing deep learning-based human mesh reconstruction approaches have atendency to build larger networks in order to achieve higher accuracy.Computational complexity and model size are often neglected, despite being keycharacteristics for practical use of human mesh reconstruction models (e.g.virtual try-on systems). In this paper, we present GTRS, a lightweightpose-based method that can reconstruct human mesh from 2D human pose. Wepropose a pose analysis module that uses graph transformers to exploitstructured and implicit joint correlations, and a mesh regression module thatcombines the extracted pose feature with the mesh template to reconstruct thefinal human mesh. We demonstrate the efficiency and generalization of GTRS byextensive evaluations on the Human3.6M and 3DPW datasets. In particular, GTRSachieves better accuracy than the SOTA pose-based method Pose2Mesh while onlyusing 10.2% of the parameters (Params) and 2.5% of the FLOPs on the challengingin-the-wild 3DPW dataset. Code will be publicly available.

 

Quick Read (beta)

loading the full paper ...