Research map / 3D vision / Depth and stereo
AMB3R: Accurate Feed-forward Metric-scale 3D Reconstruction with Backend
Hengyi Wang, Lourdes Agapito
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 sparse 3D voxel backend (Point Transformer v3) on a frozen VGGT front-end (ablation) | AMB3R · ETH3D 3D reconstruction | rel [cm]: 6.02 -> 4.64 | Improved |
| use 3D sparse backend instead of 2D alternating-attention backend (ablation) | AMB3R · ETH3D 3D reconstruction | Completeness [cm]: 12.78 -> 9.69 | Improved |
| use 3D sparse backend instead of 2D backend (ablation) | AMB3R · DTU 3D reconstruction | rel [cm]: 0.92 -> 0.81 | Improved |
| add serialization and transformer processing (S^-1 o f o S) to backend (ablation) | AMB3R · DTU / ETH3D 3D reconstruction | rel / Completeness: ETH3D rel 4.47 vs 4.64 without; DTU rel 1.02 -> 0.81 | Mixed |
| add backend to VGGT for monocular depth | AMB3R · NYUv2 monocular depth | Rel: 3.6 -> 3.0 | Improved |
| add backend to VGGT for camera pose | AMB3R · Re10K AUC@30 | AUC@30: 81.8 -> 86.3 | Improved |
| regress per-frame metric log depth at median pixel instead of global scale difference from decoder features (ablation) | AMB3R metric scale head · RMVDB metric depth | rel / delta | Improved |
| use AMB3R-VO pipeline with plain VGGT as base model (ablation) | AMB3R-VO · TUM RGB | ATE RMSE [cm]: 5.3 (VGGT-SLAM) -> 3.6 | Improved |
| add backend on top of VGGT in AMB3R-VO (ablation) | AMB3R-VO · TUM RGB | ATE RMSE [cm]: 3.6 -> 3.1 | Improved |
| add keyframe management in AMB3R-VO (ablation) | AMB3R-VO · TUM RGB | ATE RMSE [cm]: OOM -> 3.1 | Improved |
| add max pose threshold, backward search and first-frame resampling in AMB3R-VO (ablation) | AMB3R-VO · TUM RGB | ATE RMSE [cm]: 3.4 / 3.8 / 3.7 -> 3.1 | Improved |
| add global map in AMB3R-SfM (ablation) | AMB3R-SfM · ETH3D SfM | RRA@5: 94.9 -> 98.2 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- VGGT — VGGT: Visual Geometry Grounded Transformer (CVPR 2025) · better on 11 of 11 shared results
- MoGe — MoGe: Unlocking Accurate Monocular Geometry Estimation for Open-Domain Images with Optimal Training Supervision (CVPR 2025) · better on 2 of 3 shared results
- MUSt3R — MUSt3R: Multi-view Network for Stereo 3D Reconstruction (CVPR 2025) · better on 2 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
- MUSt3R: Multi-view Network for Stereo 3D Reconstruction (CVPR 2025 · similarity 0.91)
- SLAM3R: Real-Time Dense Scene Reconstruction from Monocular RGB Videos (CVPR 2025 · similarity 0.89)
- DUSt3R: Geometric 3D Vision Made Easy (CVPR 2024 · similarity 0.88)
- MASt3R-SLAM: Real-Time Dense SLAM with 3D Reconstruction Priors (CVPR 2025 · similarity 0.88)
- Pow3R: Empowering Unconstrained 3D Reconstruction with Camera and Scene Priors (CVPR 2025 · similarity 0.88)
- Fast3R: Towards 3D Reconstruction of 1000+ Images in One Forward Pass (CVPR 2025 · 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.
