Research Paper 2) Exploiting learnable vertex-to-vertex relations for generalization of human pose estimation and mesh reconstruction in real environments.

‍ learnable vertex-to-vertex relations for generalization of human pose estimation and mesh reconstruction in -realistic environments

Trung Tran-Quang, Cuong Than-Cao, Hai Nguyen-Thanh, Hong Hoang Si

Release Date: 11/1/2022

In this paper, we propose "MeshLeTemp", a powerful 3D human pose estimation and mesh reconstruction method from a single image. Unlike conventionally used fixed templates, the proposed method uses learnable human body mesh templates as prior information. The proposed method reflects not only the vertex-to-vertex interaction of the human body mesh, but also the person's posture estimation and body shape, enabling it to handle a wide variety of images. We also provide a detailed analysis of the proposed method and discuss its generality to unknown data.

‍Paper Details