Human4K introduces a large-scale 4K multi-view motion-capture dataset for whole-body 3D human reconstruction. It contains more than six million 4K images captured by an eight-view synchronized camera system, paired with professional Vicon motion capture and mocap-accurate SMPL-X annotations. The dataset covers 11 subjects performing complex, highly articulated, and strongly self-occluded motions. Its Motion-Retargeting and Refinement Module (MRRM) is designed to improve alignment across the full body and extremities. The paper reports consistent gains on standard benchmarks, especially for hands, feet, and depth-ambiguous limb configurations.
No heat snapshots are available in the last 24 hours.