Face Detection
Finding where faces are in an image - typically returning bounding boxes around each face.
Think of it like
Scanning a group photo and circling where each person is before you say their name.
Example
A camera app highlights everyones face with a blue box before snapping the photo. Social media software detects faces for auto-tagging. Security systems spot peoples faces in video streams.
How it actually works
Face detection is a localization task, distinct from face recognition (identifying who). Faces vary wildly in scale, pose (profile vs. frontal), lighting, occlusion (sunglasses, hair). Cascade classifiers (Haar, boosted) were fast but dated. Modern approaches: CNNs, RetinaNet, YOLOv3 fine-tuned on face data. Handles multiple faces, extreme angles, and occlusion but struggles in extreme lighting or crowds.
For product teams
Essential for photo apps, privacy (blur faces), security, and any system that needs to find and act on people.
For engineers
Datasets: WIDER Face (diverse), Pascal Face, AFW. Metrics: precision, recall, AP. Benchmarks vary by pose and lighting difficulty.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome