Challenges in Real-Life Face Recognition
summary
Examines face recognition failure modes when training data lacks certain variations like artistic makeup, heavy hairstyles and occlusions by introducing the BookClub artistic makeup face dataset and testing existing artificial neural network models. Reports that recognition accuracy drops sharply when test images contain unseen makeup/occlusions, and evaluates improving performance by augmenting training data with heavily made-up and occluded images; also demonstrates that such occlusions can both hide identity and cause spoofed misidentifications.
AI-written summary · about
related talks
26mNew Face, Who Dis? Recent Adversarial Approaches to Facial Recognition
28mFace spoofing detection: theory and practice
57mFoundations of Facial Recognition
26mVirtual U: Defeating Face Liveness Detection by Building Virtual Models...
31mFace Recognition
21mRecognizing Human features using Deep Networks
33mReal-Time Face Recognition: Training Fast and Accurate Models
51mWe are the Guardians of our Future