Egraphsen - Opportunities and perspectives of digital painter attribution

The question of the manufacturer of painted Greek vases has occupied archaeological research for more than 150 years. The digital analysis of Attic vases offers the opportunity to examine controversial practices of classical archaeology, which were previously based on expert knowledge. This is because it is not isolated criteria but characteristic combinations of features that determine the assignment of painters, which can be compiled and weighted much more easily and comprehensively with the aid of computers. In this project, in cooperation with the University of Göttingen, the assignment of the vases to painters, workshops and groups will no longer be exemplary and intuitive, but rather based on a broad data base. It is to be investigated how the criteria of attribution can be systematized, how their expressiveness can be evaluated and how the significance of the similarity networks can be historically weighted by precisely naming criteria and arguments.

Attic vase painting was of great importance from about 600 to 300 BC, whereby the variety of themes and styles were almost unlimited. However, only rarely have the Vase manufacturers signed their works with egraphsen (=has painted) or epoiesen(=has made), so that the attribution to painters and potters is not directly derived from by handwritten signatures. We want to create a data-driven stylometry for Attic vases, based on multimodal representations as pictures (vase pictures) and 2D ceramic profiles (ceramics)

In a first step a selection of vase pictures is annotated by human experts to mark figures, objects and ornaments. Based on on these annotations a deep convolutional neural network is trained. At a second step, architectures for deep convolutional neural networks will be to develop, which can predict whether two vase images from human experts were assigned to the same painter.

From the machine learning point of view, a monitored clustering / record linkage problem for images must be solved. We will specifically investigate models that make it possible, through semantic representations, to provide understandable explanations for the decisions of the model. Similar questions about pottery assignment, but also more advanced questions about typical cooperation scenarios between painters and potters can be answered by examining explanations and outliers.

Partner:
Institut für Digital Humanities - Georg-August-Universität Göttingen

Contact:
Prof. Schmidt-Thieme
Lukas Brinkmeyer