Research map / Video / Video-language understanding
RegVGGT: Sustainable Visual Geometry Grounding for Streaming via Regulated Memory
Hongbo Mao, Junjun Jiang, Youyu Chen, Jiaxin Zhang, Zhemeng Dong, Xianming Liu
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 global anchor preservation (keep first-frame KV) (ablation) | RegVGGT · ScanNet | ATE (m): 0.175 -> 0.165 | Improved |
| add global anchor preservation (keep first-frame KV) (ablation) | RegVGGT · ScanNet | RPE rot: 4.005 -> 3.930 | Improved |
| add layer-specific memory allocation (ablation) | RegVGGT · ScanNet | ATE (m): 0.176 -> 0.165 | Improved |
| add layer-specific memory allocation (ablation) | RegVGGT · ScanNet | RPE rot: 4.970 -> 3.930 | Improved |
| add head-specific token retention (ablation) | RegVGGT · ScanNet | Latency (ms): 17.66 -> 8.29 | Improved |
| add head-specific token retention (ablation) | RegVGGT · ScanNet | RPE rot: 3.840 -> 3.930 | Hurt |
| query downsampling 3x3 instead of 2x2 (ablation) | RegVGGT · ScanNet | RPE rot: 3.930 -> 4.224 | Hurt |
| query downsampling 4x4 instead of 2x2 (ablation) | RegVGGT · ScanNet | RPE rot: 3.930 -> 4.260 | Hurt |
| query downsampling 2x2 instead of 1x1 (ablation) | RegVGGT · ScanNet | Latency (ms): 17.64 -> 8.29 | Improved |
| lower retention rate gamma below 0.01 (ablation) | RegVGGT · ScanNet | tracking accuracy | Hurt |
| raise retention rate gamma above 0.1 (ablation) | RegVGGT · ScanNet | tracking accuracy / peak memory | No clear effect |
| replace fixed budget in Evict3R with per-frame regulation | Evict3R · Bonn video depth | Abs Rel / delta<1.25 | Improved |
Closest papers by meaning
- FrameVGGT: Coherence-Preserving Memory for Bounded Streaming Geometry (NeurIPS 2026 · similarity 0.90)
- LiteVGGT: Boosting Vanilla VGGT via Geometry-aware Cached Token Merging (CVPR 2026 · similarity 0.88)
- Rethinking the State Update Gate for Long-Sequence Recurrent 3D Reconstruction (NeurIPS 2026 · similarity 0.87)
- TurboVGGT: Fast Visual Geometry Reconstruction with Adaptive Alternating Attention (NeurIPS 2026 · similarity 0.87)
- GHOST: Geometry-Hierarchical Online Streaming Token Eviction for Efficient 3D Reconstruction (NeurIPS 2026 · similarity 0.87)
- ZPressor: Bottleneck-Aware Compression for Scalable Feed-Forward 3DGS (NeurIPS 2025 · similarity 0.86)
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.
