,

Abstract

We present YOLO, a unified pipeline for object detection that is extremely fast and accurate.

Key Contributions

  1. Unified Detection: Frame object detection as a single regression problem
  2. Real-Time Performance: 45 FPS on a Titan X GPU
  3. Full Image Context: Reasoning about global image context

Results

Model mAP FPS
YOLO 63.4 45
Fast R-CNN 70.0 0.5