CenterNet: Keypoint Triplets for Object Detection

17 Apr 2019Kaiwen DuanSong BaiLingxi XieHonggang QiQingming HuangQi Tian

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... (read more)

PDF Abstract

Evaluation results from the paper


Task Dataset Model Metric name Metric value Global rank Compare
Object Detection COCO minival CenterNet511 (Hourglass-52) box AP 41.3 # 27
Object Detection COCO minival CenterNet511 (Hourglass-52) AP50 59.2 # 23
Object Detection COCO minival CenterNet511 (Hourglass-52) AP75 43.9 # 19
Object Detection COCO minival CenterNet511 (Hourglass-52) APS 23.6 # 17
Object Detection COCO minival CenterNet511 (Hourglass-52) APM 43.8 # 19
Object Detection COCO minival CenterNet511 (Hourglass-52) APL 55.8 # 12
Object Detection COCO test-dev CenterNet511 (Hourglass-104, multi-scale) box AP 47.0 # 6
Object Detection COCO test-dev CenterNet511 (Hourglass-104, multi-scale) AP50 64.5 # 15
Object Detection COCO test-dev CenterNet511 (Hourglass-104, multi-scale) AP75 50.7 # 10
Object Detection COCO test-dev CenterNet511 (Hourglass-104, multi-scale) APS 28.9 # 9
Object Detection COCO test-dev CenterNet511 (Hourglass-104, multi-scale) APM 49.9 # 4
Object Detection COCO test-dev CenterNet511 (Hourglass-104, multi-scale) APL 58.9 # 6