Video Killed the Text Star: OSINT Approach
summary
Real-time pipeline ingests YouTube frames with OpenCV + Apache Storm, detects faces with HOG for speed and trains models offline with CNNs, producing 128‑dim face encodings from a triplet-loss network. Unknown faces are clustered across videos using Chinese whispers clustering, enabling traceability of the same person in multiple propaganda clips; demo shows knowns recognized, many unknowns grouped, and clear limits from false positives/negatives and poor training sets.
AI-written summary · about
chapters (8)
- 0:00Introduction and demo
- 3:53Open source intelligence
- 6:18Video analysis example
- 7:43Facial recognition workflow
- 21:30Technical architecture
- 26:03Grouping unknown subjects
- 29:33Advanced recognition and results
- 33:45Social impact and conclusion
related talks
21mLeveraging Python for Real-Time Image processing
58mPushing Deepfakes to the Limit-Fake video calls with AI
26mNew Face, Who Dis? Recent Adversarial Approaches to Facial Recognition
46mHow I built the world’s most efficient deepfake detector with $100
40mNew Face, Who Dis? Protecting Privacy in a World of Surveillance
01h 04mPushing Deepfakes to The Limit - Fake Video Calls With AI
51mPlaying Offense and Defense with Deepfakes
57mFoundations of Facial Recognition