Research map / 3D vision / Human pose estimation
WinT3R: Window-Based Streaming Reconstruction with Camera Token Pool
Zizun Li, Jianjun Zhou, Yifan Wang, Haoyu Guo, Wenzheng Chang, Yang Zhou, Haoyi Zhu, Junyi Chen, Chunhua Shen, Tong He
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 camera token pool over all historical windows (ablation) | WinT3R · Tanks and Temples pose (from scratch) | AUC@30: 8.87 -> 15.73 | Improved |
| add camera token pool (ablation) | WinT3R · 7-Scenes pose (from scratch) | AUC@30: 11.54 -> 15.01 | Improved |
| add camera token pool (ablation) | WinT3R · CO3Dv2 pose (from scratch) | AUC@30: 38.10 -> 47.17 | Improved |
| add camera token pool (ablation) | WinT3R · NRGBD (from scratch) | Overall: 0.350 -> 0.258 | Improved |
| add camera token pool (ablation) | WinT3R · 7-Scenes (from scratch) | Overall: 0.163 -> 0.161 | No clear effect |
| add sliding window over frame-by-frame input (ablation) | WinT3R · 7-Scenes (from scratch) | Overall: 0.212 -> 0.161 | Improved |
| add sliding window (ablation) | WinT3R · NRGBD (from scratch) | Overall: 0.404 -> 0.258 | Improved |
| add sliding window (ablation) | WinT3R · 7-Scenes pose (from scratch) | AUC@30: 7.39 -> 15.01 | Improved |
| add overlap between adjacent windows (stride w/2) (ablation) | WinT3R · 7-Scenes (from scratch) | Overall: 0.195 -> 0.161 | Improved |
| add overlap between adjacent windows (ablation) | WinT3R · CO3Dv2 pose (from scratch) | AUC@30: 44.31 -> 47.17 | Improved |
| add overlap between adjacent windows (ablation) | WinT3R · 7-Scenes pose (from scratch) | AUC@30: 11.54 -> 15.01 | Improved |
| use WinT3R vs StreamVGGT on Sintel depth | WinT3R · Sintel video depth | Abs Rel: 0.374 vs 0.343 | Hurt |
Later papers that compared with this method
2 later accepted papers report Wint3R as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- Geometric Context Transformer for Streaming 3D Reconstruction (ECCV 2026 · better on 5 of 5 shared results)
- PosePlaner: Denoising Any Feedforward Pose Predictor from Pairwise Planar Geometry (NeurIPS 2026 · better on 1 of 1 shared result)
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.
- Point3R — Point3R: Streaming 3D Reconstruction with Explicit Spatial Pointer Memory (NeurIPS 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
- Rethinking the State Update Gate for Long-Sequence Recurrent 3D Reconstruction (NeurIPS 2026 · similarity 0.88)
- CLoSeR: Closing the Loop for Long-Context Streaming Reconstruction (NeurIPS 2026 · similarity 0.86)
- BAT3R: Robust Online 3D Reconstruction with Bayesian Adaptive State Updates (NeurIPS 2026 · similarity 0.86)
- SLAM3R: Real-Time Dense Scene Reconstruction from Monocular RGB Videos (CVPR 2025 · similarity 0.85)
- Point3R: Streaming 3D Reconstruction with Explicit Spatial Pointer Memory (NeurIPS 2025 · similarity 0.85)
- OnlineSplatter: Pose-Free Online 3D Reconstruction for Free-Moving Objects (NeurIPS 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.
