Research map / 3D vision / Human pose estimation
VoxelKP: A Voxel-based Network Architecture for Human Keypoint Estimation in LiDAR Data
Jian Shi, Peter Wonka
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add spatially aware BEV fusion to VoxelNeXt baseline (ablation) | VoxelNeXt baseline · Waymo keypoints | MPJPE (all): 0.1612 -> 0.1304 | Improved |
| add sparse selective kernels (ablation) | baseline + spatial BEV · Waymo keypoints | MPJPE (all): 0.1304 -> 0.1118 | Improved |
| add sparse box-attention (ablation) | baseline + BEV + SSK · Waymo keypoints | MPJPE (all): 0.1118 -> 0.1039 | Improved |
| add hybrid feature learning (per-voxel MLP branch) (ablation) | baseline + BEV + SSK + attention · Waymo keypoints | MPJPE (all): 0.1039 -> 0.0887 | Improved |
| use direct 3D feature maps instead of spatially aware BEV (ablation) | VoxelKP · Waymo keypoints | MPJPE (all): 0.0887 -> 2.4422 | Hurt |
| increase BEV channels from 256 to 384 to compensate projection loss (ablation) | spatially aware BEV · Waymo keypoints | MPJPE (all): 0.1053 -> 0.0887 | Improved |
| use local box-attention instead of no attention (ablation) | VoxelKP · Waymo keypoints | MPJPE (all): 0.1181 -> 0.0887 | Improved |
| use stratified attention instead of box attention (ablation) | VoxelKP · Waymo keypoints | MPJPE (all): 0.0887 -> 0.1266 | Hurt |
| use stratified attention instead of no attention (ablation) | VoxelKP · Waymo keypoints | MPJPE (all): 0.1181 -> 0.1266 | Hurt |
| use VoxelKP trained only on Waymo instead of HUM3DIL | LiDAR keypoint estimation · Waymo keypoints | MPJPE (cm): 12.21 -> 8.87 | Improved |
Closest papers by meaning
- VoxelNet: End-to-End Learning for Point Cloud Based 3D Object Detection (CVPR 2018 · similarity 0.90)
- VoxelNeXt: Fully Sparse VoxelNet for 3D Object Detection and Tracking (CVPR 2023 · similarity 0.89)
- Voxel or Pillar: Exploring Efficient Point Cloud Representation for 3D Object Detection (AAAI 2024 · similarity 0.88)
- LSK3DNet: Towards Effective and Efficient 3D Perception with Large Sparse Kernels (CVPR 2024 · similarity 0.87)
- GeoAuxNet: Towards Universal 3D Representation Learning for Multi-sensor Point Clouds (CVPR 2024 · similarity 0.86)
- Pre-training a Density-Aware Pose Transformer for Robust LiDAR-based 3D Human Pose Estimation (AAAI 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
