Materials In Paintings (MIP): An interdisciplinary dataset for perception, art history, and computer vision

TL;DR

The Materials In Paintings (MIP) dataset provides a large, annotated collection of painterly depictions of materials, enabling multidisciplinary research into perception, art history, and computer vision.

Contribution

This paper introduces a novel dataset of 19,000 paintings with detailed material annotations, facilitating cross-disciplinary analysis and advancing understanding of material depiction and perception in art and AI.

Findings

Paintings depict materials using stylized approaches.

The dataset reveals distribution patterns of materials in art.

Paintings can improve robustness of computer vision models.

Abstract

A painter is free to modify how components of a natural scene are depicted, which can lead to a perceptually convincing image of the distal world. This signals a major difference between photos and paintings: paintings are explicitly created for human perception. Studying these painterly depictions could be beneficial to a multidisciplinary audience. In this paper, we capture and explore the painterly depictions of materials to enable the study of depiction and perception of materials through the artists' eye. We annotated a dataset of 19k paintings with 200k+ bounding boxes from which polygon segments were automatically extracted. Each bounding box was assigned a coarse label (e.g., fabric) and a fine-grained label (e.g., velvety, silky). We demonstrate the cross-disciplinary utility of our dataset by presenting novel findings across art history, human perception, and computer vision. Our experiments include analyzing the distribution of materials depicted in paintings, showing how painters create convincing depictions using a stylized approach, and demonstrating how paintings can be used to build more robust computer vision models. We conclude that our dataset of painterly material depictions is a rich source for gaining insights into the depiction and perception of materials across multiple disciplines. The MIP dataset is freely accessible at https://materialsinpaintings.tudelft.nl