Heart to Heart and Eye to Eye with Deep Fakes
summary
Uses pulse-derived PPG and gaze geometry as authenticity signals to detect and attribute portrait deep fakes, not just pixel artifacts. Extracts remote PPG maps from face regions, trains a tiny 3-layer CNN for in-the-wild classification and a VGG19 backbone for source ID, reporting 99.39% pairwise separation, 96.89% general detection and ~93.4% source attribution on their in-the-wild dataset, and releases a curated deep-fake-in-the-wild dataset called FakeCatcher.
AI-written summary · about
chapters (7)
- 0:00Introduction
- 1:50Deep Fakes
- 10:00Source Detection
- 15:34Eye to Eye
- 18:50Questions
- 26:13Heart to Heart
- 27:13Individual Detection
related talks
58mPushing Deepfakes to the Limit-Fake video calls with AI
01h 04mPushing Deepfakes to The Limit - Fake Video Calls With AI
51mPlaying Offense and Defense with Deepfakes
27mTruth or Fiction The Art of Unveiling DeepFakes in Text, Audio, Image
08mAdvancements in facial recognition
28mFace spoofing detection: theory and practice
51mFaceswap: Developing a platform for practical Deepfakes
16mThe Threat of Deepfakes to Remote Identity Verification Systems