Fraunhofer Institute of Optronics, System Technologies and Image Exploitation IOSB

Logos in the Wild Dataset

The Logos in the Wild Dataset is a large-scale set of web collected images with provided logo annotations in Pascal VOC style. It was created at IOSB to assess open set logo retrieval applications. The focus is on a larger brand variety than comparable in-the-wild datasets. The current version (v2.0) of the dataset consists of 11,054 images with 32,850 annotated logo bounding boxes of 871 brands.

It is an in-the-wild logo dataset where images include the logos as natural part instead of the raw original logo graphics. Images are collected by Google image search based on a list of well-known brands and companies. Brand or company names are searched directly or in combination with a predefined set of search terms, e.g., ‘advertisement’, ‘building’, ‘poster’ or ‘store’. Annotations distinguish between textual and graphical logos as well as different logos from one company as indicated in the example below.

Further statistics:

  • 4 to 608 images per searched brand
  • 238 brands occur at least 10 times
  • up to 118 logos in one image

The download link and password can be requested by sending an email to andreas.specker@iosb.fraunhofer.de

 Logos in the Wild samples

Please cite the following paper if you use the dataset in your work:

• Tüzkö A., Herrmann C., Manger D., Beyerer J.: “Open Set Logo Detection and Retrieval“, Proceedings of the 13th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications: VISAPP, 2018.

Bibtex:

@INPROCEEDINGS{,
author = {T{\"u}zk{\"o}, Andras and Herrmann, Christian and Manger, Daniel and J{\"u}rgen Beyerer},
title = {{O}pen {S}et {L}ogo {D}etection and {R}etrieval},
booktitle = {Proceedings of the 13th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications: VISAPP},
year = {2018}}