Objects as points

May 2019

tl;dr: Object detection as detection of the center point of the object and regression of its associated properties. CenterNet is the first real time anchor-free object detector.

Overall impression

CenterNet is a very generic object detection framework that can be used for 2D object detection, 3d object detection (from monocular RGB image), key point regression. The backbone can be chosen to meet different speed/accuracy tradeoff points.

FCOS regressed distances to four edges, while CenterNet only regresses width and height. The FCOS formulation is more general as it can handle amodal bbox cases (the object center may not be the center of bbox).

A quick summary of CenterNet monocular 3D object detection.

Key ideas

Technical details