ImageOP-Crack

ImageOP-Crack: The image dataset for crack detection in Brazilian cultural heritage buildings using deep learning

The ImageOP-Crack is a new public dataset for crack detection in historical buildings.

ImageOP-Crack contains 1,795 high-resolution images (4068 x 3456 pixels) with over 4,000 annotated crack labels, based on photographs collected from cultural heritage churches in the state of Minas Gerais, Brazil.

Data Collection: the data collection process was carried out between May 17 and August 10, 2025. The main objective was to capture images of cracks present on the facades of historical buildings. For this purpose, a Kodak Pixpro AZ255 digital camera was used to photograph 24 religious buildings across four historic Brazilian cities:

1) Ouro Preto: UNESCO World Heritage Site since 1980.
2) São João del-Rei: IPHAN National Heritage Site since 1938.
3) Tiradentes: IPHAN National Heritage Site since 1938.
4) Itapecerica: Municipal Heritage Site since 2006.

The photographed historical buildings preserve Brazilian colonial architecture developed during the 18th and 19th centuries. It is important to highlight that the conservation of these Brazilian cultural heritage buildings is a relevant factor, with these historic cities recognized as UNESCO World Heritage Site (Ouro Preto), National Heritage Sites (São João del-Rei and Tiradentes), and Municipal Heritage Site (Itapecerica). Notably, the main historic city photographed was Ouro Preto (OP), with 16 churches, hence, the dataset is named ImageOP-crack.

Data Annotation: subsequently, the images were annotated using the Edge Impulse platform. For each image, labels were assigned, ranging from no label (no crack), one label (one crack), or multiple labels (several cracks in the photograph).

The data are divided into two folders for training and testing deep learning systems. Additionally, the dataset includes a file with annotations for all 4,120 cracks, named info.labels.