RIPE++: Reinforced Keypoint Learning from Positive Pairs Only
zhenjun_zhao · x · 2026-08-21
The paper presents RIPE++, a method for reinforced keypoint learning using only positive pairs. Unlike traditional RL approaches like RIPE that rely on binary rewards and negative pairs, RIPE++ exploits geometric consistency signals to derive both reward and penalty from a single positive pair. This rich signal enables learning discriminative detectors and descriptors without negative contrasting. The method also extends to the matching stage, improving AUC@5 on MegaDepth1500 from 56.58 to 59.65.
More from Research
- Researcher Joins Burning Glass Institute to Focus on Economics of AI — soumitrashukla9 · 2026-08-22
- AI-assisted math lowers bound for de Bruijn–Newman constant to 0.1787854 — ctjlewis · 2026-08-22
- Machine learning studies techniques that don't work in theory but do in practice — pmddomingos · 2026-08-22
- VGB Algorithm Fixes Catastrophic Error Compounding in Long-Form Generation — arena · 2026-08-22
- Study Confirms: Asking LLMs for Concise Outputs Actually Saves Money — ibubbles34 · 2026-08-22
- BrowseComp-Plus: New Benchmark for Deep-Research Agent Evaluation — lintool · 2026-08-22