MoD: Distilling Six Detectors with Reward-Guided Routing for Keypoint Detection (NeurIPS 2026)
zhenjun_zhao · x · 2026-10-21
MoD (NeurIPS 2026) distills six heterogeneous keypoint detectors (DaD, DeDoDe, SP, ALIKED, DeepLSD, MuGE) into a unified model so diverse detection maps are predicted in one forward pass. A lightweight patch-level router then assigns the best detector per region conditioned on local texture and keypoint budget (512–8192), trained with task-driven rewards from downstream matching quality rather than heuristics. Across MegaDepth, ScanNet and HPatches and all budgets, MoD consistently beats strong baselines (e.g., AUC@5° of 68.5 on MegaDepth at budget 8192), with routing maps showing adaptive per-texture detector selection.
More from Research
- Domingos Cites New Result Claiming RL Is 'Basically Useless' vs Supervised Learning — pmddomingos · 2026-10-23
- Wolfram Opens Rule 30 Prizes to AIs: $30,000 for Three Open Problems — cephaloform · 2026-10-23
- Stanford and Databricks use AI agents to reproduce medical findings and test analytical robustness — ChengleiSi · 2026-10-23
- EMNLP paper: diffusion LMs internally encode latent denoising timestep representations — gsarti_ · 2026-10-23
- KU Leuven and VIB open two Group Leader professorships in AI for Biology, start 2027 — jpillowtime · 2026-10-23
- Subspace Tuning curbs LLM catastrophic forgetting by freezing principal singular directions — scottniekum · 2026-10-23