Program
Challenges
Cross-view Multi-object Tracking in DIVerse Open Scenes
Introduction
Cross-view multi-object tracking is a challenging problem in computer
vision that involves tracking multiple objects of interest across multiple
camera views. The goal is to associate object detections across different
views and time frames, and to maintain the identity of each object
throughout the tracking process. Interested participants are invited to
apply their approaches and methods on a novel cross-view multi-object
tracking dataset DIVOTrack being made available by the challenge
organizers. We collect data in 15 different realworld scenarios, including
indoor and outdoor public scenes. The dataset is split into development data
with a training set, testing set, and challenge set. Each of them contains
5 different real-world scenarios, including indoor and outdoor public scenes.
The duration of each video is about 1 minute at 30FPS. All the sequences are
captured by using three moving cameras and are manually synchronized. There
are both moving dense crowds and sparse pedestrians in outdoor scenes. The
surrounding environment of outdoor scenes is diverse, including streets, vehicles,
buildings, and public infrastructures. Meanwhile, the indoor scene comes from
a large shopping mall, with a more complicated and severe occlusion of the
crowd than the outdoor environment. The development data includes original
video clips, object bounding boxes, and global id for each object. The organizers
will support the evaluation and scoring the result of the challenge set. The
results of each scene has 3 txt files. The txt file format is frame_id
person_id
Ix
ly
w
h
, fid is frame id, pid is
person id, Ix is the x coordinate of the lefttop position,
ly is the y coordinate of the lefttop position, w and h are the width and height of each person box. The participants
can download the DIVOTrack from
here, and use the evaluation protocol from here
The rules for participation
The competition is open to everyone. But the members from the teams of the organizers cannot join. All teams should mailed the license for using DIVOTrack to shengyuhao (shengyuhao@zju.edu.cn). The top 3 teams will be invited to give a presentation, including methods and experimental results.
Note: Please use your institutional email to register for the competition, otherwise the registration will not be approved.
The criteria that will be used to evaluate the submitted entries
The evaluation should be user-friendly and convenient for participants. It should also be fair and safe to be hacked. We designed detailed rules as follows:
- We will limit the number of submissions each day to 1.
- Run submission files should be emailed directly to shengyuhao (shengyuhao@zju.edu.cn). The ranking will be updated on the scoreboard.
- The top 3 teams in the final scoreboard need to send their programs to the organizers. The programs are being run to reproduce their results.
The team that will run the challenge
The team running the challenge from Zhejiang University - University of Illinois Urbana-Champaign Institute. The team would be responsible for creating the challenge dataset, defining the evaluation metrics and rules, and providing support and guidance to the participants throughout the challenge.
Important dates
- Baseline release: Available now
- Training and Testing dataset release: Available now, 2023 [Available now after submitting the data agreement from located HERE]
- Challenge set release: May 15, 2023
- Submissions of solutions to organizers: October 24, 2023
- Competition results announcement: November 1, 2023
- Grand Challenge at AVSS 2023: TBD