Efficient Object Detection

Efficient object detection studies how to find and locate objects accurately while using limited computation, memory, time, and energy, especially on mobile and edge devices. Common approaches include compact YOLO and real-time DETR models, streamlined feature pyramids, pruning, quantization and teacher-to-student distillation, as well as sparse or adaptive processing that focuses computation on likely objects or informative frames. Work is tested on benchmarks such as COCO for general detection and VisDrone for aerial small objects, alongside measurements on hardware such as phones, embedded boards and FPGAs. The central challenge is balancing accuracy—particularly for tiny or difficult objects—with speed and energy use, while ensuring benchmark gains translate to reliable performance on real devices and varied scenes.