In Depth
Object detection goes beyond image classification by not only identifying what objects are present in an image but also where they are located. The model outputs bounding boxes (rectangular regions) around each detected object along with a class label and confidence score. An image might contain multiple objects of different types, all detected simultaneously.
Object detection architectures fall into two main categories: two-stage detectors like Faster R-CNN (which first propose candidate regions, then classify them) and single-stage detectors like YOLO and SSD (which predict boxes and classes in a single pass). Two-stage detectors tend to be more accurate while single-stage detectors are faster. Modern detectors increasingly use transformer-based approaches like DETR and DINO.
Object detection is among the most commercially important computer vision tasks. Applications include autonomous driving (detecting pedestrians, vehicles, signs), retail analytics (tracking customer behavior, shelf monitoring), security and surveillance (detecting people, vehicles, objects of interest), manufacturing (defect detection on production lines), and agriculture (counting fruits, detecting pests). Real-time object detection on edge devices is a rapidly growing deployment scenario.