Rocaille Segmentation Dataset

This is the dataset introduced and described in the paper "A Multimodal Dataset of 18th-Century Prints for Segmentation and Analysis of Rocaille Ornaments" (Hudcovic et al.):

We present a multimodal dataset for the computational study of 18th-century Rococo ornamental prints. The Rocaille, the era's defining ornament, exhibits a continuously transforming morphology that has historically resisted formal description. The dataset comprises 1,611 high-resolution digitizations from major German collections, accompanied by expert-curated bilingual metadata and commentary describing iconography and ornament morphology. A subset of 229 images includes pixel-accurate segmentation masks distinguishing the structural volutes and comb-like extensions of the Rocaille. This combination of imagery and detailed annotations presents several challenges for machine learning research: morphologically complex segmentation targets, substantial class imbalance, and multimodal data enabling cross-modal integration. To demonstrate the viability and utility of the dataset, we pre-train encoder–decoder architectures using diffusion-based self-supervised learning and fine-tune them for the demanding Rocaille segmentation task, providing model weights for full reproducibility. This dataset establishes a foundation for cross-disciplinary research linking art history, digital humanities, and computer vision.

Due to file limit reasons, the top-level folders were zipped. Each ZIP-file in the dataset corresponds to a top level folder, as detailed in the image folder_structure.jpg.Our paper describing and showing the viability of our dataset can be found here: [TODO]

The associated code and neural networks can be found at our codebase: https://github.com/hudo259/Rocaille-Segmentation-Experiments-FrameworkThe associated experiment artifacts (model weights, logs, visualizations) can be found at Hugging Face: https://huggingface.co/hudo259/rocaille-segmentation-experimentsCitation:[TODO]

Additionally, if you use this dataset, please also credit the following institutions from whose collections the data has been digitized:

Image courtesy of the following collections: Staatsgalerie Stuttgart, Graphische Sammlung; Staats- und Stadtbibliothek Augsburg, Graphische Sammlung; Staatliche Graphische Sammlung München.