We organize the first large-scale Tiny Object Detection (TOD) challenge, which is a competition track: tiny person detection. For this track, we will provide 1610 images with 72651 box-level annotations. We provide 18433 normal person boxes and 16909 dense boxes in training set. These image are collected from real-world scenarios based on UAVs. The persons in TinyPerson are quite tiny, and the aspect ratio of persons in TinyPerson has a large variance. Since the various poses and viewpoints of persons in TinyPerson, it brings more complex diversity of the persons. There are some annotation rules: in TinyPerson, there are only one class "person". There are four conditions where persons are labeled as “ignore”: 1) Crowds, which are hard to separate one by one when labeled with standard rectangles; 2) Ambiguous regions, which are hard to clearly distinguish whether there is one or more persons; 3) Reflections in Water, and 4) Some objects are hard to be recognized as human beings, we directly labeled them as “ignore” too. When evaluating on test set, we mainly follow pedestrian detection rules. The test set contains 13787 person boxes and 1989 ignore regions in 786 images. This challenge will be released soon, to enable participants to evaluate their approaches.
In addition to the testing results, participants are encouraged to submit a paper to RLQ workshop illustrating the challenge methodology and results. All submissions will be handled electronically via the CMT system. By submitting a paper, the authors agree to the policies stipulated on this website. TOD challenge calls for papers of two categories: Contributed Papers, and Extended Abstracts. All submissions should be in PDF format. All accepted submissions across the two categories will be invited to give poster presentations. From those accepted, a select few will be invited to give spotlight talks.
Contributed Papers follow a double-blind review process and all the accepted papers will be published as part of the "ECCV Workshop Proceedings" and should, therefore, follow the same guideline as the main conference. Paper submission guidelines of ECCV can be accessed here. Extended Abstracts follows single-blind review, and can be up to 4 pages in length (references excluded). This option is meant to provide a non-archival submission option for previously published work, or work that is intended to be published at a future venue. Supplementary appendices are allowed but will be read at the discretion of the reviewers.
Participants are required to submit a supplementary material in .zip format via firstname.lastname@example.org, including the source code, method explanation, and detailed team information etc. Details will appear soon.
Xuehui Yu, Yuqi Gong, Nan Jiang, Qixiang Ye, Zhenjun Han, Scale Match for Tiny Person Detection, the Winter Conference on Applications of Computer Vision, 2020.