Research map / 3D vision / 3D generation
STream3R: Scalable Sequential 3D Reconstruction with Causal Transformer
Yushi LAN, Yihang Luo, Fangzhou Hong, Shangchen Zhou, Honghua Chen, Zhaoyang Lyu, Bo Dai, Shuai Yang, Chen Change Loy, Xingang Pan
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 RNN state memory (CUT3R) with causal decoder-only attention over cached features (ablation) | STream3R-alpha · Sintel video depth | Abs Rel: 0.598 -> 0.535 | Improved |
| replace RNN state memory with causal decoder-only attention (ablation) | STream3R-alpha · KITTI video depth | Abs Rel: 0.157 -> 0.141 | Improved |
| replace RNN state memory with causal decoder-only attention (ablation) | STream3R-alpha · 7-Scenes 3D reconstruction | Acc mean: 0.480 -> 0.328 | Improved |
| replace RNN state memory with causal decoder-only attention (ablation) | STream3R-alpha · training throughput | training steps in same time: 60% more training steps | Improved |
| use sliding window attention (window 5) with constant cache (ablation) | STream3R-beta · KITTI video depth | FPS: 12.95 -> 32.93 | Improved |
| use sliding window attention (window 5) with constant cache (ablation) | STream3R-beta · Bonn video depth | Abs Rel: 0.069 -> 0.064 | Improved |
| use sliding window attention (window 5) with constant cache (ablation) | STream3R-beta · Sintel video depth | Abs Rel: 0.264 -> 0.279 | Hurt |
| replace causal attention with full attention at inference (ablation) | STream3R-beta · 7-Scenes 3D reconstruction | Comp mean: 0.110 -> 0.075 | Improved |
| initialize from VGG-T instead of DUSt3R weights | STream3R · Sintel single-frame depth | Abs Rel: 0.350 -> 0.228 | Improved |
| remove single-view datasets from training (ablation) | STream3R-beta · training data | performance | No clear effect |
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.
- VGG-T — VGGT: Visual Geometry Grounded Transformer (CVPR 2025) · better on 3 of 4 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
- Streaming Visual Geometry Transformer (ICLR 2026 · similarity 0.93)
- Point3R: Streaming 3D Reconstruction with Explicit Spatial Pointer Memory (NeurIPS 2025 · similarity 0.89)
- BAT3R: Robust Online 3D Reconstruction with Bayesian Adaptive State Updates (NeurIPS 2026 · similarity 0.89)
- Continuous 3D Perception Model with Persistent State (CVPR 2025 · similarity 0.88)
- Fin3R: Fine-tuning Feed-forward 3D Reconstruction Models via Monocular Knowledge Distillation (NeurIPS 2025 · similarity 0.87)
- IGGT4D: Streaming 4D Instance-Grounded Geometry Transformer (NeurIPS 2026 · similarity 0.87)
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.
