Research map / 3D vision / Depth and stereo
Back on Track: Bundle Adjustment for Dynamic Scene Reconstruction
Weirong Chen, Ganlin Zhang, Felix Wimbauer, Rui Wang, Nikita Araslanov, Andrea Vedaldi, Daniel Cremers
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 motion decoupling with the dynamic tracker (ablation) | BA-Track · MPI Sintel | ATE: 0.137 -> 0.065 | Improved |
| add camera mask (zero weight on dynamic points) to decoupled tracker (ablation) | BA-Track · MPI Sintel | ATE: 0.065 -> 0.034 | Improved |
| add camera mask to total-motion tracker (ablation) | BA-Track · MPI Sintel | ATE: 0.137 -> 0.047 | Improved |
| use dual-network decoupling instead of single static-motion tracker (ablation) | BA-Track · MPI Sintel | ATE: 0.072 -> 0.034 | Improved |
| use decoupling without mask instead of static-only tracker without mask (ablation) | BA-Track · MPI Sintel | ATE: 0.091 -> 0.065 | Improved |
| add depth consistency loss to refinement (ablation) | BA-Track · Bonn crowd2 | Abs Rel: 0.121 -> 0.103 | Improved |
| add scene rigidity loss alone to refinement (ablation) | BA-Track · Bonn crowd2 | Abs Rel: 0.121 -> 0.117 | Improved |
| add both depth and rigidity losses (ablation) | BA-Track · Bonn crowd2 | Abs Rel: 0.121 -> 0.089 | Improved |
| add depth consistency loss accuracy (ablation) | BA-Track · Bonn crowd2 | delta<1.25 (%): 89.6 -> 95.0 | Improved |
| add rigidity loss alone, accuracy (ablation) | BA-Track · Bonn crowd2 | delta<1.25 (%): 89.6 -> 88.4 | Hurt |
| apply BA-Track refinement to ZoeDepth (ablation) | ZoeDepth · MPI Sintel | Abs Rel: 0.467 -> 0.408 | Improved |
| apply BA-Track refinement to ZoeDepth (ablation) | ZoeDepth · AirDOS Shibuya | Abs Rel: 0.571 -> 0.299 | 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.
- MonST3R — MonST3R: A Simple Approach for Estimating Geometry in the Presence of Motion (ICLR 2025) · better on 3 of 5 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
- Spatiotemporal Bundle Adjustment for Dynamic 3D Reconstruction (CVPR 2016 · similarity 0.92)
- ProDyG: Progressive Dynamic Scene Reconstruction via Gaussian Splatting from Monocular Videos (NeurIPS 2025 · similarity 0.88)
- MegaSaM: Accurate, Fast and Robust Structure and Motion from Casual Dynamic Videos (CVPR 2025 · similarity 0.87)
- Enhancing 3D Reconstruction for Dynamic Scenes (NeurIPS 2025 · similarity 0.87)
- SLAM3R: Real-Time Dense Scene Reconstruction from Monocular RGB Videos (CVPR 2025 · similarity 0.87)
- 4D3R: Motion-Aware Neural Reconstruction and Rendering of Dynamic Scenes from Monocular Videos (NeurIPS 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.
