Optimized SuperPoint hits 2.1x faster keypoint detection on Jetson Orin Nano
sdiazlor · reddit · 2026-10-02
PrunaAI released an optimized version of SuperPoint for edge-device keypoint detection, up to 2.1x faster on Jetson Orin Nano.
- The approach structurally prunes the most expensive convolutional layers, recovers performance via distillation, and accelerates keypoint selection with hierarchical top-k, preserving the original architecture's core behavior.
- The distilled model retains strong keypoint coverage across indoor and outdoor data with low descriptor differences from the original.
- Weights are open on Hugging Face, and the optimizations are applicable to other runtimes.
More from Research
- NYU's DAGent: evaluate-then-grow planning lifts deep research agents by up to 5.8 points — newyorkuniversity · 2026-10-02
- Amazon's MILO auto-evolves agent harnesses, hitting 86.1% on Terminal-Bench 2.1 — amazon · 2026-10-02
- Prompted identity splits multi-agent LLM teams: success drops from 96% to 81% — Xavier Del Giudice · 2026-10-02
- Microsoft's Rubric Response Theory beats GRPO by up to 5.6 points with half the judge budget — microsoft · 2026-10-02
- New SI-cpWER benchmark tests persistent speaker identity across meetings; ThyVoice leads at 47.13 — Shantanu Vispute · 2026-10-02
- Sonia Joseph co-founds World Mechanics, a lab building interpretable foundation models for physical world — soniajoseph_ · 2026-10-02