Research map / Architectures and efficiency / Attention and transformers
OVGGT: O(1) Constant-Cost Streaming Visual Geometry Transformer
Si-yu Lu, Po-Ting Chen, Hui-Che Hsu, Sin-Ye Jhong, Wen-Huang Cheng, Yung-Yao Chen
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 Dynamic Anchor Protection (initial + historical anchors) vs none (ablation) | OVGGT · KITTI depth 500 frames, far | F1% gain over no-anchor: +5.43% (initial only) -> +10.15% (full) | Improved |
| add historical anchors on top of initial anchor (ablation) | OVGGT · KITTI depth 500 frames, near | F1% gain: +3.51% -> +5.35% | Improved |
| increase max historical anchors beyond 3 (ablation) | OVGGT · KITTI depth 500 frames | RMSE (near): 1.546 (K=3) vs 1.552 (K=5), 1.548 (K=10) | No clear effect |
| increase cache budget 100K -> 200K (ablation) | OVGGT · 7-Scenes 300 frames | Acc (mean): 0.059 -> 0.026 | Improved |
| increase cache budget 200K -> 1M (ablation) | OVGGT · 7-Scenes 300 frames | Acc (mean): 0.026 -> 0.028 | No clear effect |
| balance hybrid scoring beta 0.1 -> 0.5 (ablation) | OVGGT · 7-Scenes 300 frames | Acc (median): 0.021 -> 0.009 | Improved |
| add activation smoothing alpha 0.0 -> 0.5 (ablation) | OVGGT · 7-Scenes 300 frames | Acc (mean): 0.027 -> 0.026 | No clear effect |
| FFN residual scoring instead of Q.K dot product (ablation) | OVGGT · 7-Scenes 300 frames | Acc (mean): 0.032 -> 0.026 | Improved |
| FFN residual scoring vs attention-weight scoring (ablation) | OVGGT · 7-Scenes 300 frames | Acc (mean): 0.023 (attention) vs 0.026 (FFN) | Mixed |
| bounded cache with selection vs full cache (StreamVGGT) | OVGGT · KITTI video depth | Abs Rel: 0.166 -> 0.128 (100 frames) | Improved |
| increase budget to 400K | OVGGT · ETH3D | Acc (mean): 0.628 -> 0.535 | Improved |
Closest papers by meaning
- FrameVGGT: Coherence-Preserving Memory for Bounded Streaming Geometry (NeurIPS 2026 · similarity 0.94)
- GHOST: Geometry-Hierarchical Online Streaming Token Eviction for Efficient 3D Reconstruction (NeurIPS 2026 · similarity 0.91)
- IGGT4D: Streaming 4D Instance-Grounded Geometry Transformer (NeurIPS 2026 · similarity 0.90)
- LiteVGGT: Boosting Vanilla VGGT via Geometry-aware Cached Token Merging (CVPR 2026 · similarity 0.90)
- TurboVGGT: Fast Visual Geometry Reconstruction with Adaptive Alternating Attention (NeurIPS 2026 · similarity 0.89)
- Rethinking the State Update Gate for Long-Sequence Recurrent 3D Reconstruction (NeurIPS 2026 · 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.
