TROGS-26: The ICDAR 2026 Competition in Text Recognition on Greek Squeezes
Overview
Paper squeezes record and preserve written text inscribed in stone or other media. Squeeze collections are currently being scanned and digitized at scale, opening up an important source of historical information to further study. This competition seeks to identify the best recognition methods for use with squeezes, based upon a selection of scanned squeezes that were originally collected from museums and archaeological sites in the eastern Mediterranean.
Rules
One entry is allowed per person/team. For planning purposes and to receive relevant updates, teams considering an entry in the competition should register their interest. Registration does not imply any obligation to continue.
The contest is organized as a single track competition with transcription character error rate as the evaluation criterion. Test images will be released one week prior to the submission deadline, at which point training and development ceases and no changes to the method are allowed. All contestants must certify that they have not used the test images to alter their method, including any form of training or parameter setting. The organizers of the competition will rely on the academic integrity of the participants when reporting final results.
Results will be written up for publication following the conclusion of the contest. Top entries will be asked to provide an outline of their approach suitable for publication in the contest report. Winners will be formally presented at the ICDAR 2026 conference.
Dataset
The IAS squeezes contain a wide variety of texts such as laws, honorific decrees, contracts, epitaphs, dedications, and even poetry. With origins spanning the ancient Mediterranean from Athens to Anatolia, these texts often address subjects and ideas that fall outside the purview of contemporary literary histories.
Training data provided for this contest consists of 224 annotated squeezes. Each squeeze has been scanned under two orthogonal lighting conditions, for a total of 448 annotated images. Annotations are available in PageXML. Download the training data here.
Test data will consist of additional, previously unpublished annotated squeezes from the collection. The full dataset will be made available following the conclusion of the event.
Code to compute the error rate will be forthcoming.
Test images will be released on or around March 27, 2026.
Schedule
- December 1, 2025
- Competition website is live; training set available
- March 27, 2026
- Test images released; no alteration of entries from this point onwards
- April 3, 2026
- Competition submission deadline
- Aug 31- Sep 2, 2026
- Formal presentation of results at ICDAR conference
Team
Nicholas R. Howe
Smith College
Aaron Hershkowitz
Institute for Advanced Study
Contact
For questions and general inquiries about this competition, please email nhowe@smith.edu.