QMUL-OpenLogo: Open Logo Detection Challenge

Hang Su         Xiatian Zhu         Shaogang Gong

Computer Vision Group, School of Electronic Engineering and Computer Science, Queen Mary University of London

Open-set 1 logo detection results

Method All Class Unsupervised Class Supervised Class
YOLO9000 [1] 4.19 1.98 26.33
YOLOv2 [1]+SCL [2] 12.10 8.75 45.48
YOLOv2 [1]+CAL [4] 13.14 9.55 49.17
FR-CNN [3]+SCL [2] 12.35 8.51 50.74
FR-CNN [3]+CAL [4] 13.13 9.34 51.03

Open-set 2 logo detection results

Method All Class Unsupervised Class Supervised Class
YOLOv2 [1]+SCL [2] 30.06 12.75 47.36
YOLOv2 [1]+CAL [4] 30.16 13.72 46.60
FR-CNN [3]+SCL [2] 33.89 18.63 49.16
FR-CNN [3]+CAL [4] 34.25 20.31 48.19

Closed-set logo detection results

Method mAP
FR-CNN [3] 48.3

Model Download

YOLOv2 [1]+SCL [2]

FR-CNN [3]+SCL [2]

YOLOv2 [1]+CAL [4]

FR-CNN [3]+CAL [4]

YOLOv2 config file

FR-CNN train prototxt file

FR-CNN test prototxt file


[1]. Jpseph Redmon and Ali Farhadi. Yolo9000: better, faster, stronger. CVPR 2017.

[2]. Hang Su, Xiatian Zhu and Shaogang Gong. Deep learning logo detection with data expansion by synthesising context. WACV 2017.

[3]. Shaoqing Ren, Kaiming He, Ross Girshick, and Jian Sun. Faster r-cnn: Towards real time object detection with region proposal networks. NIPS 2015.

[4]. Hang Su, Xiatian Zhu and Shaogang Gong. Open Logo Detection Challenge. BMVC 2018.