Research map / 3D vision / Depth and stereo
Geometric Context Transformer for Streaming 3D Reconstruction
Lin-Zhuo Chen, Jian Gao, Shangzhan Zhang, Yihang Chen, Nan Xue, Jianyuan Wang, Christian Rupprecht, Xun Cao, Xing Zhu, Yujun Shen, Yao Yao, YINGHAO XU
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 anchor initialization (ablation) | GCT baseline · TartanGround val | AUC@3: 9.80 to 13.63; ATE 8.59 to 7.88 | Improved |
| add relative pose loss (ablation) | GCT · TartanGround val | ATE: 8.25 -> 7.46; RPE-rot 5.35 -> 2.26 | Improved |
| add video RoPE on top of context tokens (ablation) | GCT · TartanGround val | ATE: 7.46 -> 5.98; AUC@3 15.75 -> 16.39 | Improved |
| add context tokens (trajectory memory) (ablation) | GCT · TartanGround val | AUC@30: 68.25 -> 69.92 | Improved |
| increase window size from 16 to 64 (ablation) | GCT · Oxford Spires | ATE: 11.2 -> 5.3 (n=8) | Improved |
| increase window size from 64 to 128 (ablation) | GCT · Oxford Spires | ATE: 5.5 -> 5.6 (n=2); 5.3 -> 5.3 (n=8) | No clear effect |
| increase anchor count n from 2 to 8 (ablation) | GCT · Oxford Spires | ATE: 5.5 -> 5.3 (k=64) | No clear effect |
| go from sparse (320 frames) to dense (3,840 frames) sequences | GCT vs CUT3R · Oxford Spires | ATE: GCT 5.37 -> 5.60; CUT3R 18.16 -> 32.47 | Improved |
| paged KV-cache with FlashInfer instead of contiguous PyTorch cache | GCT · 518x378 video inference | FPS: ~10.5 FPS -> ~20 FPS | Improved |
| replace causal attention with GCA | GCT · context size, T=10,000 | tokens in context: ~5x10^6 -> ~7x10^4 | 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.
- Wint3R — WinT3R: Window-Based Streaming Reconstruction with Camera Token Pool (ICLR 2026) · better on 5 of 5 shared results
- TTT3R — TTT3R: 3D Reconstruction as Test-Time Training (ICLR 2026) · better on 2 of 2 shared results
- VGGT — VGGT: Visual Geometry Grounded Transformer (CVPR 2025) · 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
- CLoSeR: Closing the Loop for Long-Context Streaming Reconstruction (NeurIPS 2026 · similarity 0.87)
- IGGT4D: Streaming 4D Instance-Grounded Geometry Transformer (NeurIPS 2026 · similarity 0.87)
- GHOST: Geometry-Hierarchical Online Streaming Token Eviction for Efficient 3D Reconstruction (NeurIPS 2026 · similarity 0.86)
- Point3R: Streaming 3D Reconstruction with Explicit Spatial Pointer Memory (NeurIPS 2025 · similarity 0.86)
- FrameVGGT: Coherence-Preserving Memory for Bounded Streaming Geometry (NeurIPS 2026 · similarity 0.86)
- GeoLRM: Geometry-Aware Large Reconstruction Model for High-Quality 3D Gaussian Generation (NeurIPS 2024 · 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.
