VASC: training-free sparse attention speeds up 3D reconstruction inference by up to 2.29x
zhenjun_zhao · x · 2026-10-02
A new arXiv paper introduces VASC (Value-Aware Sparse Attention with Cross-Layer Memory), a training-free method addressing the quadratic cost of global attention in feed-forward 3D vision models like VGGT:
- Value-aware block selection combines pooled query–key relevance with neighboring value contrast to cut redundancy while keeping query-relevant, distinctive content.
- Cross-layer memory tracks unserved demand across layers and updates based on actual execution, letting underserved blocks compete under a fixed compute budget.
On 7Scenes and NeuralRGB-D with VGGT and π³, it beats FasterVGGT on pose estimation and reconstruction quality, with up to 2.29× faster inference than dense VGGT. Code is released.
More from Research
- Marin's 535B MoE hero run hits ~27% MFU on 11 NVL72 racks: expert parallelism deep dive — dlwh · 2026-10-03
- Harvard-led team unveils brain imaging 60x faster than fMRI, tracking activity in ~100ms — melnykowycz · 2026-10-02
- RExBench: best coding agent implements research extensions only 33% of the time — najoungkim · 2026-10-02
- Retrieval-augmented episodic memory narrows LLMs' lexical frequency gap in syntax tests — najoungkim · 2026-10-02
- EMPIRIC teaches robots missing physics as code, solving all 25 tasks where baselines get 14-16 — tomssilver · 2026-10-02
- What the AI pain paper actually found, according to someone who read it — Drukarshar · 2026-10-02