chcaa/CHR_detection_trees

Motif Segmentation in Paintings

Segmentation masks and metadata for trees detected in a corpus of paintings, produced with SAM3 as part of research in the golden matrix project at the Center for Humanities Computing Aarhus (chcaa).

Dataset Description

This dataset supports a studies of landscape and tree motifs in 19th-century Northern European paintings. It covers 4,727 paintings drawn from museum and Wikidata sources, with tree instances detected using [SAM3]

For each detected tree, the dataset records a bounding box, a confidence score, and a segmentation mask, alongside painting-level metadata (artist, nationality, date, collection)

Contents

Each row corresponds to one painting and includes:

• filename, image — original image filename and raw image bytes
• title, artist, Artist-nationality, inception, collection, location, inventory_number, qid, wikidata_url, image_url_raw, image_filename — painting metadata
• Image-rights — rights/licensing information for the original image, as recorded per painting
• original_height, original_width — original image dimensions
• boxes, scores, count — detected tree bounding boxes, confidence scores, and count
• embedding — image embedding vector
• mask — raw bytes of an .npz file; load with np.load(io.BytesIO(row["mask"]))["masks"] to get a boolean array of shape (count, 1, H, W)

License

This dataset (masks, metadata, embeddings, and annotations produced by the chcaa team (Rie Schmidt Eriksen, Louise Brix Hansen & Marta Kipke)) is released under CC-BY-4.0. Please credit authors when reusing this data.