Research map / 3D vision / Depth and stereo
TALO: Pushing 3D Vision Foundation Models Towards Globally Consistent Online Reconstruction
Fengyi Zhang, Tianjun Zhang, Kasra Khosoussi, Zheng Zhang, Zi Huang, Yadan Luo
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 |
|---|---|---|---|
| replace Sim(3) global alignment with TPS alignment and pose-averaging registration | VGGT with 5 cameras · Waymo | RRE RMSE (deg): 0.71 -> 0.14 | Improved |
| replace Sim(3) alignment with TALO | VGGT, pi3, MapAnything · Waymo ATE avg | ATE RMSE (m): VGGT 1.42 -> 1.09; pi3 2.22 -> 0.86; MapAnything 3.68 -> 1.40 | Improved |
| replace 15-DOF SL(4) alignment with 7-DOF Sim(3) or TALO | 3D foundation models in outdoor multi-camera settings · Waymo and nuScenes | ATE RMSE (m): VGGT-SLAM 12.21 -> TALO 1.09 (VGGT, Waymo) | Improved |
| replace Sim(3) alignment with TALO on nuScenes | VGGT, pi3, MapAnything · nuScenes ATE avg | ATE RMSE (m): VGGT 1.63 -> 1.31; pi3 1.63 -> 1.02; MapAnything 2.34 -> 0.91 | Improved |
| replace Sim(3) alignment with TALO | VGGT, pi3, MapAnything · Waymo | Chamfer Distance: VGGT 0.75 -> 0.62; pi3 0.81 -> 0.59; MapAnything 1.27 -> 1.07 | Improved |
| replace Sim(3) alignment with TALO | VGGT, pi3, MapAnything · nuScenes | Chamfer Distance: VGGT 1.09 -> 1.11; pi3 1.22 -> 1.17; MapAnything 1.31 -> 1.10 | Mixed |
| replace Sim(3) alignment with TALO in monocular front-camera setting | VGGT · Waymo monocular | Chamfer Distance: 1.96 -> 1.96 | No clear effect |
| replace Sim(3) alignment with TALO in monocular front-camera setting | VGGT · Waymo monocular | ATE RMSE (m): 1.27 -> 1.08 | Improved |
| increase number of cameras (1, 3, 5) (ablation) | VGGT-Long and TALO · Waymo | Chamfer Distance | Improved |
| increase submap size (2 to 8) (ablation) | VGGT-Long and TALO · Waymo | ATE RMSE (m) | Improved |
| increase number of cameras and submap size (ablation) | VGGT-SLAM (SL(4)) · Waymo | ATE and Chamfer | Improved |
| use TALO instead of online DUSt3R-like methods CUT3R and MASt3R-SLAM | monocular Waymo · Waymo monocular | ATE RMSE (m): MASt3R-SLAM 3.92 -> 1.08; CUT3R 9.44 -> 1.08 | 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
- Online3R: Online Learning for Consistent Sequential Reconstruction Based on Geometry Foundation Model (CVPR 2026 · similarity 0.88)
- FoundationGeo: Learning Spatial Pixel-Wise Fields for Monocular Metric Geometry (ECCV 2026 · similarity 0.88)
- Tango3D: Towards Alignment for Global and Local 2D-3D Correspondence (NeurIPS 2026 · similarity 0.87)
- PF3plat: Pose-Free Feed-Forward 3D Gaussian Splatting for Novel View Synthesis (ICML 2025 · similarity 0.86)
- Fin3R: Fine-tuning Feed-forward 3D Reconstruction Models via Monocular Knowledge Distillation (NeurIPS 2025 · similarity 0.85)
- LoRA3D: Low-Rank Self-Calibration of 3D Geometric Foundation models (ICLR 2025 · similarity 0.85)
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.
