Reproduce “LiveHPS: LiDAR-based Scene-level Human Pose and Shape Estimation in Free Environment” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Yiming Ren, Xiao Han, Chengfeng Zhao, Jingya Wang, Lan Xu, Jingyi Yu, Yuexin Ma

LiveHPS estimates full-body pose, shape, and global location from LiDAR point-cloud sequences. It aligns point-cloud features with a body-model reference, refines joints across time and body parts, then predicts SMPL parameters. On FreeMotion, it reports joint/vertex errors of 119.27/128.61 mm, versus 129.88/141.77 mm for LIP’s LiDAR-only branch under the paper’s comparison setup; occlusion and noise remain limitations.

Key insights

  • Aligning diverse LiDAR point cloud features with regular SMPL mesh vertices via a Vertex-guided Adaptive Distillation (VAD) mechanism effectively regularizes point distributions for improved pose and shape estimation.
  • Leveraging temporal and spatial attention mechanisms within a Consecutive Pose Optimizer (CPO) enables robust pose refinement by exploiting motion coherence and geometric joint relationships, even with incomplete and noisy LiDAR data.
  • A Skeleton-aware Translation Solver can accurately estimate global human translation by predicting a bias, effectively addressing the instability and error accumulation issues of prior methods.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.