Leveraging the Learnable Vertex-Vertex Relationship to Generalize Human Pose and Mesh Reconstruction for In-the-Wild Scenes
This work addresses the problem of robust human reconstruction in varied real-world settings for computer vision applications, representing an incremental advance over prior methods.
The paper tackles 3D human pose and mesh reconstruction from single images by introducing a learnable template mesh that adapts to diverse scenes, achieving improved generalizability on unseen scenarios as demonstrated through extensive experiments.
We present MeshLeTemp, a powerful method for 3D human pose and mesh reconstruction from a single image. In terms of human body priors encoding, we propose using a learnable template human mesh instead of a constant template as utilized by previous state-of-the-art methods. The proposed learnable template reflects not only vertex-vertex interactions but also the human pose and body shape, being able to adapt to diverse images. We conduct extensive experiments to show the generalizability of our method on unseen scenarios.