Code Submissions¶
Inputs and Outputs for Each Task
Task 1: Action Recognition and Action Anticipation:
- The input for the action recognition and action anticipation tasks will be images representing the frames of each video in the dataset. The output would be two action class names, one for the recognized action and the other for the anticipated action, represented as a string.
- Sample Algorithm: https://github.com/zhuoyp/T3Challenge2025/tree/main/sample_algorithms/actions
Task 2: Hand Tracking
- The input is the frames of each of the videos. The output should include the ID of the hand and the coordinates of the bounding box for each hand in YOLO format.
- Sample Algorithm: https://github.com/zhuoyp/T3Challenge2025/tree/main/sample_algorithms/hands
Task 3: Object Detection
- The input is the frames of each of the videos. The output should include the name of the object as a string and the coordinates for the bounding box for each object in YOLO format.
- Sample Algorithm: https://github.com/zhuoyp/T3Challenge2025/tree/main/sample_algorithms/objects
Task 4: Realism Assessment
- The inputs are the frames of each of the videos. The output should include the realism score as an integer.
- Sample Algorithm: https://github.com/zhuoyp/T3Challenge2025/tree/main/sample_algorithms/realism
Task 5: Visual Question Answering
- The inputs are images extracted from the videos in the original dataset, and the output should be a string containing the answer to the question.
- Sample Algorithm: https://github.com/zhuoyp/T3Challenge2025/tree/main/sample_algorithms/vqa
Algorithms will be assessed online automatically through the Grand Challenge website. To submit results, participants can use the provided Docker template, upload their results to the designated portal on the website, and give time for the evaluation to work. After the evaluation is finished, the results will automatically appear on the website, but they are allowed to request a withdrawal at any time.
The participants are allowed multiple submissions for all tasks, and the last run is officially counted as the final result. To avoid test set leakage, the total number of submissions allowed during the testing phase is 2. It includes any failed submissions due to incorrect formats.
Paper Submissions¶
There are 5 tasks in total for the challenge. All accepted papers will be included in the MICCAI challenge proceedings (Lecture Notes in Computer Science (LNCS) volume in the challenges subline). To be eligible for prizes, it is mandatory to submit a paper with a minimum of 8 pages. The paper should cover at least one of the five tasks. If you are participating in multiple tasks, you have the option to submit a single paper that includes all methods and results, or you can submit several papers.
Authors should follow Springer's authors' guidelines and utilize their proceedings templates, either LaTeX or Word. Springer encourages authors to include their ORCIDs (Open Researcher and Contributor IDs) in their papers. Additionally, the corresponding author, representing all the co-authors, must complete and sign a Consent-to-Publish form. It is important that the corresponding author indicated on the copyright form matches the corresponding author in the paper. Please note that once the files have been submitted, no change to the authorship of the papers can be made. To submit the Consent-to-Publish form, please append it to the end of the paper.
All paper submissions should uploaded to the CMT submission platform*: https://cmt3.research.microsoft.com/TTTC2025.
The proceeding from the Trauma THOMPSON Challenge 2023 is available here: https://link.springer.com/book/10.1007/978-3-031-71626-3
We strongly urge participants to share their code and include the GitHub link in their papers.
The best performing teams that have submitted papers will be contacted and requested to prepare an oral presentation for the challenge satellite event during MICCAI 2025.
Participants are also welcome to submit their papers to conferences and journals after the conclusion of the challenge.
*The Microsoft CMT service was used for managing the peer-reviewing process for this conference. This service was provided for free by Microsoft and they bore all expenses, including costs for Azure cloud services as well as for software development and support.