Thanks to the intensive work carried out by our colleagues at TU Wien, the LEGION project is pleased to announce the release of its first joint publication, together with an accompanying open-access benchmark dataset: CENTURIA.
The CENTURIA dataset provides 507 retro-digitised Roman pottery documentation records from Carnuntum, complete with handwritten text transcriptions, bounding boxes, and structured field-level metadata across seven categories (provenance, project, find number, stratigraphic unit, pottery form, fabric, publication type, and rim diameter). The accompanying paper evaluates five state-of-the-art OCR models and introduces an expert-validated LoRA fine-tuning workflow that reduces transcription error to below 1.5% and achieves over 87% field extraction accuracy.
The paper will be presented next week by our colleagues from TU Wien at ECCV 2026 (European Conference on Computer Vision) in Malmö, Sweden, in two renowned workshops: Women in Computer Vision (WiCV) and Vision for Art and Culture (VISART VIII).
The work is currently available as an open-access preprint on arXiv, with the final version to appear in the official conference proceedings (Springer LNCS) in approximately one year.
Paper: OCR-Based Field Extraction for Archaeological Pottery Metadata: The CENTURIA Dataset
Gissu Valentina Naghavi, Dominik Hagmann, Martin Kampel, Irene Ballester
Preprint / DOI: https://doi.org/10.48550/arXiv.2608.30616
Dataset & Code (GitHub): https://github.com/gissuvalentina/CENTURIA
This marks the first major milestone and dataset release from LEGION (machine LEarninG-enabled Identification of archaeological Objects in the middle daNube river basin), a project dedicated to the automated classification of Roman pottery using AI-based methods, funded by Heritage Science Austria 2.0 under the Austrian Academy of Sciences (OeAW).
For more information about the project, visit the LEGION website.
LEGION is based at Austrian Archaeological Institute.
PI: Dominik Hagmann (OeAI–OeAW). Team: Silvia Radbauer (OeAI–OeAW), Irene Ballester Campos (TU Wien), Martin Kampel (TU Wien), Sebastian Zambanini (TU Wien).
Cooperation partners: Lower Austrian State’s Collections, University for Continuing Education Krems, Archaeological Park Carnuntum, OeAW–ACDH (ARCHE), Vienna Museum