The dataset TopLogo-10 contains 10 unique logo classes related to most popular brands of clothing, shoes, and accessories. Each class has 70 images collected from the Flickr website, therefore providing realistic challenges for automated logo detection algorithms. All the 700 logo images are manually labelled with fine-grained object bounding boxes.


We benchmark two data partitions (A & B) as below.

Partition Classes Tranining Images per Class (Total) Test Images per Class (Total)
A 10 40 (400) 30 (300)
B 10 10 (100) 60 (600)

Icons of 10 logo classes




Hang Su, Xiatian Zhu, Shaogang Gong.
Deep Learning Logo Detection with Data Expansion by Synthesising Context
IEEE Winter Conference on Applications of Computer Science (WACV), Santa Rosa, USA, March 2017. 
arxiv bibtex  


For any queries, please contact Hang Su at


Please notice that this dataset is made available for academic research purpose only. All the images are collected from the Internet, and the copyright belongs to the original owners. If any images belong to you and you would like them to be removed, please kindly inform us.