CenterNet: Keypoint Triplets for Object Detection

TL;DR

CenterNet introduces a novel keypoint triplet approach for object detection, enhancing accuracy and speed by detecting objects as triplets of keypoints and employing specialized pooling modules, outperforming existing detectors on MS-COCO.

Contribution

The paper proposes CenterNet, a new object detection framework that models objects as triplets of keypoints, with novel pooling modules, improving detection precision and recall over prior keypoint-based methods.

Findings

Achieves 47.0% AP on MS-COCO, outperforming existing one-stage detectors.

Provides faster inference speed with comparable performance to two-stage detectors.

Introduces cascade corner pooling and center pooling modules for better feature representation.

Abstract

In object detection, keypoint-based approaches often suffer a large number of incorrect object bounding boxes, arguably due to the lack of an additional look into the cropped regions. This paper presents an efficient solution which explores the visual patterns within each cropped region with minimal costs. We build our framework upon a representative one-stage keypoint-based detector named CornerNet. Our approach, named CenterNet, detects each object as a triplet, rather than a pair, of keypoints, which improves both precision and recall. Accordingly, we design two customized modules named cascade corner pooling and center pooling, which play the roles of enriching information collected by both top-left and bottom-right corners and providing more recognizable information at the central regions, respectively. On the MS-COCO dataset, CenterNet achieves an AP of 47.0%, which outperforms all existing one-stage detectors by at least 4.9%. Meanwhile, with a faster inference speed, CenterNet demonstrates quite comparable performance to the top-ranked two-stage detectors. Code is available at https://github.com/Duankaiwen/CenterNet.