Photo sleuth: combining human expertise and face recognition to identify historical portraits
summary
A web platform called Photo Sleuth combines crowdsourced expertise and automated face recognition to identify American Civil War portraits. It seeds the database with 17,000 seeded photos, asks volunteers to add visual tags mapped to complete military records, and uses a three-stage “build, narrow, compare” haystack pipeline. Launched in August, the first month yields ~600 users, ~2,000 uploads and 119 new identifications (perfect matches in easy cases, 25/37 correct for hard cross-view cases), showing face recognition + crowd is effective but imperfect.
AI-written summary · about
chapters (32)
- 0:00Introduction
- 0:29Motivations
- 1:02Creating economic value
- 1:25Recognition of contributions
- 2:04Personal connections
- 3:04American Civil War
- 4:57Uniform
- 5:35Reference materials
- 6:30Tax stamp
- 6:56Additional photos
- 7:15Website
- 7:47Website workflow
- 8:06Add photo
- 8:25Visual tags
- 8:40Military records
- 9:28Search results
- 10:31haystack problem
- 11:00narrow down haystack
- 11:28comparison view
- 11:47volunteers
- 12:00inperson events
- 12:24media coverage
- 12:45analysis
- 13:49how did it help
- 15:09fun quotes
- 17:07Private Collections
- 17:29Current Status
- 17:46Takeaways
- 18:13Sustainable contributions
- 18:35Complementary strengths
- 18:58Questions
- 24:57Future work
related talks
50mCabbage, or recognizing 330 million faces at 1000 photos/sec
26mVirtual U: Defeating Face Liveness Detection by Building Virtual Models...
57mFoundations of Facial Recognition
26mNew Face, Who Dis? Recent Adversarial Approaches to Facial Recognition
38mVideo Killed the Text Star: OSINT Approach
01h 18mComputer Vision after the Victory of Data
25mPersonalizing Your Images with AI-Powered Features
24mChallenges in Real-Life Face Recognition