How To Submit¶

For detailed information about the opening and closing dates of the submission phases, please visit the ODIN 2025 webpage.

The challenge repository will publish the evaluation scripts, baseline algorithms, input/output specifications, and example submission templates for both tasks.

Please always refer to the structured submission document for detailed information.

General Submission Policy¶

The test datasets will not be publicly released. Participants must submit either a Docker container or a GitHub repository following the templates provided by the organizers.

All submitted algorithms must be fully automatic. No user interaction is allowed during evaluation.

Participants are not allowed to submit multiple solutions under different team names or accounts. Submissions violating this rule will be removed from the final ranking and the associated participants will be disqualified.

Allowed Training Data and Models¶

Participants may use publicly available data and publicly available models, provided that all data and models were available before challenge submissions open.

Participants must describe all datasets, models, pre-training sources, and external annotations used in their method description. Private data are not allowed unless made publicly available to the research community before the submission phase.

Code Release Requirement¶

The top three performing methods must share their code in a public repository, such as GitHub, with a README that enables replication of the submitted method. If a winning team does not release its code in a timely manner, the prize will be awarded to the next eligible institution or research group.

To be considered for the post-challenge publication, teams must make their algorithm publicly available, including network parameters when applicable, under an appropriate license.

Task 1 - ToothFairy4: CBCT to Maxillofacial/Surgical Report¶

Input: A submission receives one CBCT case at a time. Each case contains a pre-processed 3D CBCT volume of the jaws.
Output: The algorithm must output a textual clinical report for the case.

Debugging Phase¶

During the Debugging Phase, submissions are evaluated on a subset of cases extracted from the training set. This phase is intended to test Docker execution, input/output compliance, and report formatting. During this phase, the evaluation will compute a subset of the overall metrics. Results from this phase will not contribute to the final ranking.

Test Phase¶

During the Test Phase, submissions are evaluated on 50 hidden CBCT cases. The hidden test set comes from an independent center and will not be released. Teams are allowed two final submissions. The best submission will be used for the final ranking.

Since some evaluation metrics cannot be computed directly on the grand-challenge platform (e.g., RadFact), all submitted solutions will be re-evaluated locally.

The ranking is explained in the Ranking page.

Task 2 - Bite2Text: IOS and Intraoral Photographs to Orthodontic Report¶

Input: A submission receives one multimodal case at a time. Each case contains upper and lower intraoral 3D scans and the associated set of standardized intraoral photographs.
Output: The algorithm must output an orthodontic report for the case.

Debugging Phase¶

Test Phase¶

During the Test Phase, submissions are evaluated on 50 hidden patient cases. The hidden test set comes from centers not included in the training data and will not be released. Teams are allowed two final submissions. The best submission will be used for the final ranking.

Since some evaluation metrics cannot be computed directly on the grand-challenge platform (e.g., RadFact), all submitted solutions will be re-evaluated locally.

The ranking is explained in the Ranking page.