End-to-End Recovery of Human Shape and Pose

TLDR

Unlike most methods that compute only 2D or 3D joint locations, we produce a richer mesh representation parameterized by shape and 3D joint angles, but the reprojection loss alone is highly underconstrained. The study introduces Human Mesh Recovery (HMR), an end‑to‑end framework that reconstructs a full 3D mesh from a single RGB image by minimizing reprojection loss of keypoints, enabling training on in‑the‑wild images with only 2D annotations. The method introduces an adversarial network trained on a large 3D human mesh database to discriminate realistic shape and pose parameters, and directly infers 3D pose and shape from image pixels without intermediate 2D keypoint detections. HMR can be trained with or without paired 2D‑to‑3D supervision, runs in real time given a bounding box, outperforms previous optimization‑based mesh methods on in‑the‑wild images, and achieves competitive results on 3D joint estimation and part segmentation.

Abstract

We describe Human Mesh Recovery (HMR), an end-to-end framework for reconstructing a full 3D mesh of a human body from a single RGB image. In contrast to most current methods that compute 2D or 3D joint locations, we produce a richer and more useful mesh representation that is parameterized by shape and 3D joint angles. The main objective is to minimize the reprojection loss of keypoints, which allows our model to be trained using in-the-wild images that only have ground truth 2D annotations. However, the reprojection loss alone is highly underconstrained. In this work we address this problem by introducing an adversary trained to tell whether human body shape and pose parameters are real or not using a large database of 3D human meshes. We show that HMR can be trained with and without using any paired 2D-to-3D supervision. We do not rely on intermediate 2D keypoint detections and infer 3D pose and shape parameters directly from image pixels. Our model runs in real-time given a bounding box containing the person. We demonstrate our approach on various images in-the-wild and out-perform previous optimization-based methods that output 3D meshes and show competitive results on tasks such as 3D joint location estimation and part segmentation.