ExtremeNet: Bottom-up Object Detection by Grouping Extreme and Center Points

April 2019

tl;dr: Detect four extreme corners and the center of an object with an anchor-less framework. This is heavily influenced by CornetNet, and eventually leads to CenterNet.

Overall impression

The bottom up approach treats object detection as keypoint detection and keypoint grouping. It is related to and inspired by the annotation method in extreme clicking.

Key ideas

Technical details