Robust Full Body 3D Human Pose Estimation

This Master's thesis investigates robust full-body 3D human pose estimation from monocular images. Existing unified methods often struggle to capture fine-grained body details, particularly for the hands and face.

The work proposes a composition-of-experts framework that combines specialized models within the unified SMPL-X representation. SMPLest-X is used for global body pose and shape estimation, WiLoR for detailed hand reconstruction, and EMOCA for expressive facial geometry. The pipeline aligns and fuses the outputs of these models to obtain a coherent full-body 3D mesh while preserving the strengths of each specialized component.

The work was conducted at the African Institute for Mathematical Sciences (AIMS), South Africa, in 2025.