Misanthrope: A Privacy-Preserving Keypoint Detector that avoids detecting people
zhenjun_zhao · x · 2026-08-26
Misanthrope is a privacy-preserving keypoint detector trained via self-distillation to avoid detecting keypoints on people, mitigating inversion attacks at the source. While traditional pipelines allow adversaries to reconstruct privacy-sensitive content like people from local features, Misanthrope prevents this. It maintains image matching performance on par with state-of-the-art methods and even surpasses them in challenging settings with people as distractors. The model ranked as the top-performing sparse feature extractor in 7 out of 9 scenes on the Image Matching Challenge 2021 Phototourism test set.
More from Safety
- Bill Gates warns we’ve crossed AI danger thresholds in bio, cyber, and jobs — nordicinst · 2026-08-26
- 10-25% of MATS Fellows Are Not 'Safety-Pilled', Estimate Suggests — AaronBergman18 · 2026-08-26
- Has Anyone Actually Suffered a Prompt Injection Attack? — cinooo1 · 2026-08-26
- Study Reveals SynthID Watermark Easily Detectable and Scrubable — basedjensen · 2026-08-26
- Circumventing AI watermarks directly in-prompt with PRNGs — JulianHabekost · 2026-08-26
- User documents persistent discrepancy between Gemini history and Google Activity records — Aggressive-Arm-1182 · 2026-08-26