1. You can download the test set at the 'data' page (yuncong_test_set.zip)
2. Each team can upload no more than 3 times before 2019/1/28 23:59 (UTC time). No score will be returned. The final leaderboard for test set phase will be released on Feb 5. Please note that not all photos in the test set will be scored.
1. fixed bug in the evaluation program.
2. if you have made any submission before, please DO NOT re-submit, we will re-evaluate all existed submission in the next 1-2 weeks.
We replace the evaluation algorithm due to its speed issue. Please accept our apologies and find more information below:
1, we use a new evaluation metric because the previous one occupied too much computational time and resources. The new metric is Precision-recall. Please find more details on the evaluation page.
2, If you have made any submission before, please DO NOT re-submit, we will re-evaluate all existed submission in the next 1-2 weeks.
3, the deadline of the competition is now on Jan 18, 2019.
With the continuous expansion and prosperity of urban commercial districts, more and more people are shopping, catering, finding leisure and entertainment in the business district. Behind the bustling business district is a huge hidden danger. In the event of an emergency, all types of risks will be amplified in crowd district. In recent years, many crowded and trampling accidents occurred worldwide have fired the alarm for the management of urban business districts.
Cloudwalk Technology has been incubating from the Chongqing Institute of Green and Intelligent Technology, Chinese Academy of Sciences in 2015. Cloudwalk has also helped draft and develop national standard on face recognition. Its technology application covers many industries, such as banking, security, and aviation.
In 2018, National Development and Reform Commission (NDRC) entrusted CloudWalk, alongside with Beijing Zhongdun under the First Research Institute of the Ministry of Public Security, for the construction of the key program of National AI Project, the Industrialization and Application of High Accuracy Face Recognition System.
When we want to calculate how many people in a very crowded scene, we naturally calculate the number of people based on the visible part of the body, and the visible part is mainly the head area. This drove us to detect head counting rather than the entire human-based to solve crowd counting problem. The contest invites participants to design algorithms, mainly for the detection of human heads in complex scenes, by indicating the specific position (x, y, w, h) of each person's head rectangle. In which, x represents the coordinate on the x axis, and y is the coordinate on the y axis, w is the width of the face rectangle, and h is the length of the face rectangle.
If you have any questions, please ask it in post in the discussion webpage, or send an email to email@example.com. _
2018 Cloudwalk Headcount