Yao Wei ; Zhang Hongwen ; Sun Yunlian ; Liu Yebin ; Tang Jinhui

Abstract

Previous methods for 3D human motion recovery from monocular images oftenfall short due to reliance on camera coordinates, leading to inaccuracies inreal-world applications. The limited availability and diversity of focal lengthlabels further exacerbate misalignment issues in reconstructed 3D human bodies.To address these challenges, we introduce W-HMR, a weak-supervised calibrationmethod that predicts "reasonable" focal lengths based on body distortioninformation, eliminating the need for precise focal length labels. Our approachenhances 2D supervision precision and recovery accuracy. Additionally, wepresent the OrientCorrect module, which corrects body orientation for plausiblereconstructions in world space, avoiding the error accumulation associated withinaccurate camera rotation predictions. Our contributions include a novelweak-supervised camera calibration technique, an effective orientationcorrection module, and a decoupling strategy that significantly improves thegeneralizability and accuracy of human motion recovery in both camera and worldcoordinates. The robustness of W-HMR is validated through extensive experimentson various datasets, showcasing its superiority over existing methods. Codesand demos have been made available on the project pagehttps://yw0208.github.io/w-hmr/.