Research map / 3D vision / Human pose estimation
VoxelTrack: Multi-Person 3D Human Pose Estimation and Tracking in the Wild
Yifu Zhang, Chunyu Wang, Xinggang Wang, Wenyu Liu, Wenjun Zeng
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 |
|---|---|---|---|
| Increase voxel size from 80x80x32 to 120x120x48 (ablation) | VoxelTrack · Panoptic | AP25: 39.74 -> 71.00 | Improved |
| Increase voxel size from 120x120x48 to 160x160x64 (ablation) | VoxelTrack · Panoptic | AP25: 71.00 -> 79.34 | Improved |
| Increase voxel size from 120x120x48 to 160x160x64 (ablation) | VoxelTrack · Panoptic | MPJPE: 19.83 mm -> 18.49 mm | Improved |
| Use sparse convolution instead of standard convolution (160x160x64) (ablation) | VoxelTrack · Panoptic | JEN time: 132.64 ms -> 48.46 ms | Improved |
| Use sparse convolution instead of standard convolution (80x80x32) (ablation) | VoxelTrack · Panoptic | JEN time: 18.75 ms -> 22.01 ms | Hurt |
| Use 5 cameras instead of 3 (ablation) | VoxelTrack · Panoptic | AP25: 49.09 -> 79.34 | Improved |
| Use 5 cameras instead of 3 (ablation) | VoxelTrack · Panoptic | ID Switch: 0 -> 0 | No clear effect |
| Use Higher-HRNet-W32 backbone instead of DLA-34 (ablation) | VoxelTrack · Panoptic | AP25: 79.34 -> 85.88 | Improved |
| Use MobileNet-V2 backbone instead of DLA-34 (ablation) | VoxelTrack · Panoptic | AP25: 79.34 -> 42.42 | Hurt |
| Use larger input image 960x512 instead of 640x384 (ablation) | VoxelTrack · Panoptic | AP25: 55.96 -> 79.34 | Improved |
| Add Re-ID features to 3D-pose-only tracking (ablation) | VoxelTrack · Panoptic | ID Switch: 90 -> 15 | Improved |
| Add occlusion mask when fusing Re-ID features (ablation) | VoxelTrack · Panoptic | IDF1: 94.38 -> 98.67 | Improved |
Later papers that compared with this method
1 later accepted paper reports VoxelTrack as a baseline in their results tables; 0 of them report a better number on most of the results they share with it.
- TEMPO: Efficient Multi-View Pose Estimation, Tracking, and Forecasting (ICCV 2023 · better on 0 of 2 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- TesseTrack: End-to-End Learnable Multi-Person Articulated 3D Pose Tracking (CVPR 2021 · similarity 0.91)
- Faster VoxelPose: Real-Time 3D Human Pose Estimation by Orthographic Projection (ECCV 2022 · similarity 0.90)
- VoxelPose: Towards Multi-Camera 3D Human Pose Estimation in Wild Environment (ECCV 2020 · similarity 0.90)
- Direct Multi-view Multi-person 3D Pose Estimation (NeurIPS 2021 · similarity 0.89)
- Unsupervised Cross-Modal Alignment for Multi-Person 3D Pose Estimation (ECCV 2020 · similarity 0.88)
- FreeMan: Towards Benchmarking 3D Human Pose Estimation under Real-World Conditions (CVPR 2024 · similarity 0.88)
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.
