The paper compares and mostly contrasts YOLO with other models. Those models being:
- Deformable parts models (DPM)
- R-CNN (and fast and faster R-CNN)
- Deep MultiBox
As a theme, the YOLO either pioneered or is very different in its aggregations of functions to predict from input to bounding box + class. Most other models pipeline (quite complexly) the problem, which is what I would do if I was designing an algorithm. The major advantage claimed by YOLO is the speed. But on top of speed, it still boasts the ability to make relatively high numbers of classes irrespective of apparent size and location.
YOLO boasts of its realtime speed