Research map / 3D vision / Human pose estimation
HumanBA: Human-Aware Bundle Adjustment via Global Human-Camera Decoupling
Fengyuan Yang, Tanuj Sur, Tze Ho Elden Tse, Angela Yao
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 human joints as pseudo-static BA landmarks on top of masked DROID with scale estimation (ablation) | Masked DROID-SLAM · EMDB2 | W-MPJPE (mm): 230.97 -> 195.97 | Improved |
| add human joints as BA landmarks (ablation) | Masked DROID-SLAM · EMDB2 | ATE (m): 0.369 -> 0.358 | Improved |
| add human motion refinement (jump filtering and smoothing) (ablation) | HumanBA · EMDB2 | ATE (m): 0.404 -> 0.358 | Improved |
| use all joints with equal weight instead of adaptive weighting (ablation) | HumanBA · EMDB2 | ATE (m): 0.508 vs 0.358 (full) | Hurt |
| use only the root joint as target instead of adaptive weighting (ablation) | HumanBA · EMDB2 | ATE-S (m): 0.718 vs 0.682 (full) | Hurt |
| add acceleration-based adaptive target weighting (ablation) | HumanBA · EMDB2 | WA-MPJPE (mm): 74.81 -> 70.10 | Improved |
| use no dynamic mask in DROID-SLAM (ablation) | DROID-SLAM · EMDB2 | ATE-S (m): 7.25 -> 7.957 (vs masked) | Hurt |
| evaluate HumanBA on motion-spike sequences only | HumanBA vs masked DROID + scale est. · EMDB2 (motion-spike sequences) | ATE-S (m): 0.649 -> 0.525 | Improved |
| replace TRAM with HumanBA on long sequences | camera scale-aware error · EMDB2 (long sequences) | ATE-S (m): 0.81 -> 0.90 | Hurt |
| replace masked DROID with HumanBA | camera ATE · EMDB2 (all sequences) | ATE (m): 0.34 -> 0.32 | No clear effect |
| replace TRAM with HumanBA | local pose · EMDB2 | PA-MPJPE (mm): 38.6 -> 38.6 | No clear effect |
| replace TRAM with HumanBA | global human · SLOPER4D (3-sequence subset) | ATE-S (m): 4.05 -> 3.10 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DROID-SLAM — DROID-SLAM: Deep Visual SLAM for Monocular, Stereo, and RGB-D Cameras (NeurIPS 2021) · better on 1 of 1 shared result
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
- Decoupling Human and Camera Motion From Videos in the Wild (CVPR 2023 · similarity 0.89)
- HumanMM: Global Human Motion Recovery from Multi-shot Videos (CVPR 2025 · similarity 0.89)
- WHAM: Reconstructing World-grounded Humans with Accurate 3D Motion (CVPR 2024 · similarity 0.89)
- Synergistic Global-space Camera and Human Reconstruction from Videos (CVPR 2024 · similarity 0.88)
- ODHSR: Online Dense 3D Reconstruction of Humans and Scenes from Monocular Videos (CVPR 2025 · similarity 0.87)
- Reconstructing People, Places, and Cameras (CVPR 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.
