Human4K: A Large-Scale 4K Multi-View Mocap Dataset for Whole-Body 3D Human Reconstruction

arXiv cs.AI · 18 hours ago

Researchers released Human4K, a dataset containing over six million 4K images from eight synchronized cameras and motion capture data for training 3D human reconstruction models. The dataset includes recordings of 11 subjects performing complex full-body motions with precise SMPL-X annotations covering hands, feet, and occluded body parts. Models trained on Human4K show improved performance on standard benchmarks, particularly for reconstructing hands, feet, and depth-ambiguous limb configurations.