Research map / Video / Video-language understanding
STAC: Plug-and-Play Spatio-Temporal Aware Cache Compression for Streaming 3D Reconstruction
Runze Wang, Yuxuan Song, Youcheng Cai, Ligang 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 anchor tokens selected by decayed cumulative attention (ablation) | STAC · NRGBD ablation | accuracy (lower better): 0.0725 -> 0.0648 | Improved |
| add long-term spatial token caching (ablation) | STAC · NRGBD ablation | accuracy (lower better): 0.0713 -> 0.0648 | Improved |
| add long-term spatial token caching (ablation) | STAC · NRGBD ablation | memory (GB): 0.572 -> 2.210 | Hurt |
| add count-based attention bias for merged tokens (ablation) | STAC · NRGBD ablation | accuracy (lower better): 0.0666 -> 0.0648 | Improved |
| add chunk-based multi-frame optimization (ablation) | STAC · NRGBD ablation | backbone runtime (ms): 138.12 -> 71.18 | Improved |
| add chunk-based multi-frame optimization (ablation) | STAC · NRGBD ablation | accuracy (lower better): 0.0673 -> 0.0648 | Improved |
| integrate STAC into STream3R | STream3R · NRGBD/7-Scenes | total memory (GB): 19.75 -> 2.20 | Improved |
| integrate STAC into STream3R | STream3R · NRGBD/7-Scenes | FPS: 2.52 -> 10.53 | Improved |
| integrate STAC into STream3R | STream3R · NRGBD stride 5 | accuracy (mean, lower better): 0.053 -> 0.065 | Hurt |
| use STAC instead of sliding window at matched memory | STream3R · NRGBD stride 5 | accuracy (mean, lower better): 0.088 -> 0.065 | Improved |
| integrate STAC into StreamVGGT | StreamVGGT · NRGBD stride 5 | accuracy (mean, lower better): 0.134 -> 0.126 | Improved |
| integrate STAC into StreamVGGT | StreamVGGT · Sintel pose | ATE: 0.224 -> 0.251 | Hurt |
Closest papers by meaning
- GHOST: Geometry-Hierarchical Online Streaming Token Eviction for Efficient 3D Reconstruction (NeurIPS 2026 · similarity 0.89)
- FrameVGGT: Coherence-Preserving Memory for Bounded Streaming Geometry (NeurIPS 2026 · similarity 0.86)
- Rethinking the State Update Gate for Long-Sequence Recurrent 3D Reconstruction (NeurIPS 2026 · similarity 0.86)
- Point3R: Streaming 3D Reconstruction with Explicit Spatial Pointer Memory (NeurIPS 2025 · similarity 0.85)
- BAT3R: Robust Online 3D Reconstruction with Bayesian Adaptive State Updates (NeurIPS 2026 · similarity 0.85)
- ZPressor: Bottleneck-Aware Compression for Scalable Feed-Forward 3DGS (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.
