HERITAGE SCIENCE AUSTRIA 2.0

LEGION

machine LEarninG-enabled Identification of archaeological Objects in the middle daNube river basin.

Decoding the history of Roman Carnuntum through automated classification of common ware pottery using cutting-edge Machine Learning.

EXPLORE PROJECT

Project News

Digital Frontiers Workshop

22 May 2026

Workshop: Digital Frontiers

Dominik Hagmann presented "Reflections on AI Applications in Austrian Roman Archaeology" at the Digital Frontiers workshop, highlighting LEGION's Human-in-the-Loop approach, legacy data digitization, and the transparent, ethical use of Generative AI.

LEGION Presentation at MLA2S (Photo at the MLA2S: after Martin Kampel - modified using ChatGPT 5.5)

11 May 2026

MLA2S Networking Seminar

Dominik Hagmann and Irene Ballester-Campos presented LEGION at the 9th Networking Seminar of the MLA2S platform, highlighting AI-enabled identification of archaeological objects from Carnuntum.

Event Details →
MAIA Second General Meeting in Hainburg (Photo of Hainburg by Onur Birol/MAIA - modified using ChatGPT 5.5)

7 April 2026

MAIA General Meeting

Dominik Hagmann hosted the Second General Meeting of the MAIA COST Action in Hainburg, focusing on managing AI in archaeological research through interdisciplinary collaboration.

Full Report →

Coming Soon

First Data Release

We are currently preparing the first batch of digitised archaeological drawings for public release.

Fragmented Knowledge & Data Overload

The Manual Bottleneck

Traditional classification of everyday ceramics is too slow for massive, fragmented archaeological datasets. Huge amounts of data remain unassessed.

Untapped Data Points

Tens of thousands of technical 2D find drawings remain unclassified and unintegrated due to the sheer volume of material.

Some finds are more equal than other finds

Funding for mass finds is often restricted. Automated solutions are required to bridge this gap.

AI-Enabled Archaeology

LEGION introduces a Semi-supervised Computer Vision Pipeline leveraging thousands of historical documents for deep material insight.

Hybrid Data Input (Photo of analog file folders in the OeAI/OeAW's archive and LSNÖ's archaeological central depot at Hainburg by Dominik Hagmann and LSNÖ, modified using Gemini 3.1 Pro and ChatGPT 5.5)

Hybrid Data Input

Processing over 70,000 2D drawings alongside their comprehensive archaeological metadata.

Human-in-the-Loop

Human-in-the-Loop (HITL)

eXplainable AI (XAI) ensures results are validated by experts to remain archaeologically robust.

Open Source Integration Visualization

Open Source Integration

Fully open-source solutions hosted on GitHub, ensuring long-term accessibility and transparency for the global research community.

New Socio-Economic Insights & AI Tools

Local Production & Trade Visualization

Local Production & Trade

Revealing complex patterns in local production and tracking trade trends across the entire Middle Danube River Basin.

New Typochronology Visualization

New Typochronology

Establishing a completely new, data-driven typochronology of Roman Common Ware based on large scale analysis.

High Classification Accuracy

Free & Open-Source Classification Tool

A user-friendly, open-source tool built for rapid, automated typochronological dating with >90% classification accuracy.

Experts Behind LEGION

The LEGION project is a multi-disciplinary collaboration bridging the gap between archaeological expertise and advanced computational methods.

We are always looking for collaboration and feedback. Reach out to the project lead for inquiries regarding data, methodology, or partnership opportunities.

OeAW Official Project Page @OeAI
TU Wien Official Project Page @CVL
GitHub Open Source Repository

OeAI Team

Austrian Archaeological Institute (OeAI/OeAW)

Dominik Hagmann PI
Silvia Radbauer Co-PI
Julia Tanzer Student Assistant

CVL Team

Computer Vision Lab, TU Wien

Irene Ballester Campos Co-PI
Martin Kampel Co-PI
Sebastian Zambanini Co-PI
Valentina Naghavi Master's student
Anna Laczkó Master's student

Leading Institutions