"Introduction to Face Processing with Computer Vision|"
summary
Walks through concrete pipelines and numbers: early Haar-like features and HOG + SVM give cheap real-time frontal detection, R‑CNN variants cut per-image time from minutes to ~200 ms (claim: Fast R‑CNN ≈200× faster than R‑CNN; Faster R‑CNN enables few‑hundred‑ms inference). Face recognition shifts from classification to R‑CNN → Fast R‑CNN → Faster R‑CNN metric learning with embeddings/metric learning, reaching ≈100% on LFW for good crops; WIDER FACE still splits easy/medium/hard in the 90s. Prototypes use a few lines of code to get embeddings/landmarks; cloud APIs can process ~1,000 faces per dollar.
AI-written summary · about
chapters (6)
related talks
57mFoundations of Facial Recognition
52mFace Recognition: From Scratch To Hatch
01h 12mBuilding a live face recognition system
31mFace Recognition
33mReal-Time Face Recognition: Training Fast and Accurate Models
39mFaster than real-time face detection
21mLeveraging Python for Real-Time Image processing
39mBuilding Highly Scalable Facial Recognition Pipelines