AltarData

Published: 3 August 2026| Version 1 | DOI: 10.17632/2b6sd9n5gy.1
Contributors:
,

Description

AltarData: A novel annotated dataset for object detection in the Brazilian heritage altars using deep learning AltarData innovates by publicly providing high-quality images of altars from 33 Brazilian historical religious buildings located in 12 historic towns. Three of these urban ensembles are recognized as UNESCO World Heritage Sites: the Historic Town of Ouro Preto, the Historic Centre of Salvador de Bahia, and the Historic Centre of the Town of Diamantina. In this context, the proposed dataset comprises 1,278 photographs with 12,093 labeled instances distributed across the following categories: 1 - Altar Table: the table located at the front of the altar, where items for the Eucharistic celebration are usually placed. 2 - Column: cylindrical pillar that serves both as a structural element and as an ornament of the altars. 3 - Crucifix: sculpture of Jesus Christ on the cross. In this study, both small crucifixes, typically placed in front of the tabernacle, and larger sculptures positioned at the center of the altar were considered. 4 - Saint Sculpture: images of saints located on the main or side altars of the churches. 5 - Tabernacle: small cabinet (box) with a door at the center of the altar, where the hosts are stored. 6 - Throne: type of pedestal used to display saint sculptures and crucifixes. In this study, both central and side thrones of the altars were considered. The data annotation phase aimed to label each collected altar image. The object types were selected based on their consistent presence in the altars of Brazilian historical churches from the 18th and 19th centuries. In this paper, the dataset was annotated by the authors, based on a careful study of traditional references in the field of Brazilian religious Baroque. In this sense, the architectural and ornamentation glossary of the Minas Gerais Baroque (A. Ávila, J. M. M. Gontijo, R.G. Machado, Barroco Mineiro: Glossário de Arquitetura e Ornamentação, Fundação João Pinheiro, 1980) and the visual glossary of the Luso-Brazilian Baroque (B. L .de Toledo, Esplendor do Barroco Luso-Brasileiro, 2nd Edition, Atelie Editorial, 2015) were analyzed. The dataset was split into training, validation, and test sets based on the geographical location of the historic churches: - training: Training and Validation Experiments (29 churches – 1,226 images), with images from the cities of Ouro Preto, Mariana, Serro, Rio de Janeiro, Cachoeira do Campo, Tiradentes, São João del-Rei, Lagoa Dourada, São Brás do Suaçuí, and Salvador. It is worth noting that the images in this subset were randomly split in an 80/20 ratio for training and validation, respectively. - testing: Test Experiments (4 churches – 52 images), with images from the cities of Diamantina and Campanha. In addition, the AltarData images are accompanied by an auxiliary file named info.labels. This file contains the annotations for each of the 12,093 items across all images.

Files

Institutions

Categories

Artificial Intelligence, Computer Vision, Cultural Heritage, Brazil, Deep Learning

Funders

  • Universidade Federal de Ouro Preto
    Grant ID: Edital PROPPI 18/2024

Licence