Student Topics

Feature Learning and Clustering for Archaeological Pottery Typology

Master Thesis

Status: available
Supervisors: Martin Kampel, Irene Ballester

Build AI for Archaeological Discovery at a UNESCO World Heritage Site

Work with real data from Roman Carnuntum (70,000+ pottery drawings) to create classification tools that archaeologists will actually use. Your work will contribute to ongoing Heritage Science research in Austria.

Context

This thesis tackles a real-world archaeological challenge in collaboration with the Austrian Archaeological Institute (ÖAI) and Archaeological Park Carnuntum. You’ll develop computational methods to analyze a large collection of Roman pottery drawings from Carnuntum, one of the most important Roman sites along the Danube frontier (1st-4th century CE). Archaeological pottery drawings present unique challenges: varying quality, heterogeneous styles, and diverse preservation states. You’ll work on: (1) preprocessing using domain-specific tools, (2) learning meaningful feature representations, and (3) clustering to discover coherent pottery groupings.

This work connects to the LEGION Heritage Science project, a collaboration between TU Wien’s Computer Vision Lab and the Austrian Archaeological Institute. Your methods and findings could directly inform tools used by researchers across Europe.

Tasks

 

  1. Literature review: computer vision for technical drawings andarchaeological artifacts, feature learning, clustering methods
  2. Implement and evaluate PyPotteryLens preprocessing tool; adapt for dataset-specific challenges
  3. Investigate and compare feature extraction approaches (pre-trained models, self-supervised learning, domain-specific methods)
  4. Implement and compare clustering algorithms
  5. Evaluate clustering quality through quantitative metrics and visual analysis
  6. Optional: explore integration of archaeological metadata (chronology, context, measurements…)
  7. Document methodology, experiments, and results

Deliverables

  1.  Master thesis
  2. Documented and clean code repository (preprocessing, feature extraction, clustering)
  3. Preprocessed dataset with extracted features and cluster assignments

Are You a Good Fit?

  • Strong Python programming skills (PyTorch)
  • Background in computer vision/machine learning (or eager to learn)
  • Independent and systematic working style
  • Excited about interdisciplinary research and real-world impact

What to expect

  • Duration: typically 6-8 months full-time, flexible based on availability
  • Weekly meetings to discuss progress and next steps (online or in-person)
  • Optional participation in project meetings with archaeologists and site visits to Carnuntum
  • Opportunity to contribute to research publications

Contact

Irene Ballester (irene.ballester@tuwien.ac.at)
Martin Kampel (martin.kampel@tuwien.ac.at)