The study »OCR-Based Field Extraction for Archaeological Pottery Metadata: The CENTURIA Dataset« by lead author Gissu Valentina Naghavi, Dominik Hagmann, Martin Kampel, and Irene Ballester received the Best Paper Award at the VISART Workshop of the renowned European Conference on Computer Vision (ECCV) 2026. It demonstrates how artificial intelligence can turn handwritten archaeological annotations into structured, searchable data.

The CENTURIA benchmark dataset comprises 507 selected, digitised pottery drawings from Carnuntum with expert-validated reference data (ground truth). The team compared five models for optical character recognition (OCR) and handwritten text recognition (HTR), including multimodal vision-language models (VLMs). Adapting the models using just 57 training examples manually transcribed by experts reduced the transcription error rate to below 1.5%. Comparable records can therefore be processed automatically with high accuracy, potentially reducing manual transcription work substantially.

The work forms part of LEGION, a collaboration between the Austrian Archaeological Institute of the Austrian Academy of Sciences and TU Wien’s Computer Vision Lab, funded through the Academy’s Heritage Science Austria 2.0 programme. Future work will focus on extending these methods to larger archival collections and integrating them with AI-assisted classification of Roman pottery.

Preprint: https://doi.org/10.48550/arXiv.2608.30616