Challenge Registration

Register your team to participate in this challenge by filling out the registration form on the following link. Registration deadline is October 27. 

Participation 

The challenge is open to both students and working individuals.

A. Registration Requirements

• Mandatory Registration
  All participants, including solo challengers, must complete and submit the Team Registration Form prior to joining the competition. Completing the Team Registration Google Form is sufficient for challenge registration at this stage. No additional CMT registration is required now. The ABC2027 CMT link and instructions for paper and supplementary-material submission will be announced separately.

• Team Size
  There is no restriction on the number of team members, but every team member is expected to contribute actively.

B. Team Composition

• Team Members

  • There is no limit to team size.

  • Only supervisors can participate in multiple teams within the same challenge.

• Special Considerations

  • If creating multiple teams from the same lab, ensure compliance with the No Private Sharing rule.

  • For team mergers, participants must contact the organizers via email.

• Participation Across Challenges

  • Students are allowed to join multiple ABC challenges.

However, students affiliated with the host or organizing laboratories are not eligible to participate.

C. Data Use, Training, and Code Sharing

Training data and pretrained models

Training and fine-tuning are permitted only on the official Training set provided for this competition. External training datasets, including other publicly released TrackID3x3 data and annotations outside the official Training set, are not permitted.

As an exception, publicly available general-purpose pretrained models may be used, provided they have not been trained or fine-tuned on the corresponding Challenge sequences or annotations. Any additional training or fine-tuning must use only the official Training set. Participants must disclose the model checkpoints, their sources, and the data used for additional training.

Use of the Challenge videos

The Challenge videos may be used for inference and the court calibration described below. Self-training or fine-tuning of model weights on these videos, including training on automatically generated pseudo-labels, is not permitted.

Normal online tracking updates during inference, such as updating track states, appearance-feature histories, or templates, are permitted.

Manual calibration and annotation

Manual selection of fixed court landmarks for camera/court calibration is permitted. Participants must describe any manual calibration procedures in their method report.

Manual frame-by-frame annotation or correction of player positions, bounding boxes, tracks, team labels, or jersey identities in the Challenge videos is not permitted.

Public ground-truth annotations

Top-view annotations corresponding to the Challenge sequences are publicly available outside the official competition Training set. These annotations must not be used for training or pseudo-label generation. Corresponding ground-truth trajectories must not be copied, time-aligned, replayed, or otherwise used to generate submitted predictions.

All submitted predictions must be generated from the Challenge videos by the reported automatic method.

Code and data sharing

Private sharing of competition-related code or data outside your team is prohibited, except for required submissions to the organizers and reviewers. Submission of code for integrity and reproducibility checks does not require public release of the code.

D. Challenge Completion Requirements

To successfully complete the challenge, participants must:

1. Prediction Submission: https://forms.gle/rMwK4pHm1Xp2joV27

   • Each team must submit predictions for all Challenge evaluation videos once. There is no separate intermediate Test set or Test-set ranking stage. 

   • Predictions will be collected through a dedicated Google Form. The submission link and upload instructions will be announced separately. Please provide the same team name used in the Team Registration Form.

   • Submit one prediction file per Challenge video, using the video basename: IMG_0114.txt and IMG_0115.txt.

   • Each file must contain comma-separated values without a header, with the following five fields in this order: frame_id, court_x, court_y, team_id, jersey_number

   • The field names above indicate the column order only; do not include them as a header row.

   • Use the original video frame numbering, starting at 1. Court coordinates must be in centimeters and use the same coordinate system as the Training ground truth. Team IDs must be A or B. The combination of team ID and jersey number defines the player identity; no additional numeric track-ID column is required.

   • All frames are evaluated, including frames before the game starts. Do not trim the sequences or renumber the frames.

2. Code and Model Submission

   • Please prepare a ZIP archive containing your source code, configuration files, and a README with the instructions needed to run your method and reproduce the submitted predictions.

   • Provide the model weights used for inference. For weights or other files exceeding the configured upload limit, include fixed-version download links in the README. Keep these files accessible to the organizers and reviewers until the checks are complete. Copies of the video dataset do not need to be included.

   • The submitted code, model weights, and settings must be the exact versions used to generate the predictions submitted by the November 3 deadline. Please preserve those versions at the prediction deadline; the later code submission must not describe a subsequently improved method.

At least one team member is required to present the paper in the workshop.

3. Paper Submission

   • Provide a paper detailing your methodology in alignment with the challenge objectives.

At least one team member is required to present the per in the workshop.

By default, accepted papers will be published under IJABC (International Journal of Activity Behavior and Computing) under Jstage. For paper format, please check the ABC Main website.

Evaluation Overview

Submissions are evaluated using TI-HOTA with a 100 cm court-distance parameter. TI-AssA follows the public evaluation implementation], using an unweighted mean over ground-truth identities.

All frames of each Challenge video are evaluated, including frames before the game starts. Please preserve the original video frame numbering.

Paper Submission

The paper should be written in English, following the ABC formatting guidelines. The paper should be 8-12 pages including references. The expected contents of the paper are:

• Abstract: A concise summary of the team’s proposed method and the results.

• Methodology: A detailed description of the methodology, including the dataset, preprocessing, model development, and evaluation metrics.

• Results and Discussion: A presentation and analysis of the results and a discussion of the strengths and limitations of the proposed method.

• Conclusion: A summary of the team’s main findings, implications, and suggestions for future work.

• References: A list of references, including three main papers of the challenge (published paper about the dataset). Challenge Summary Paper will be provided upon confirmation with the hosts. Kindly submit camera-ready ahead.

For information regarding the Registration for the ABC conference, visit the main ABC2027 website.

Prizes

• The winning team will be awarded 100,000 JPY and Wearable Devices for Activity and Behaviour Tracking (approximate Total Value $800 - $1000).

• Each of the participating teams will be awarded with participation certificate.