HW-Data: Heritage Window Dataset

The Heritage Window Dataset (HW-Data) consists of window images collected from Brazilian historic buildings. Initially, the data collection process involved photographing nine heritage churches between July 21 and November 9, 2025, across four Brazilian historic towns: Ouro Preto, Campanha, Mariana, and Diamantina. For this purpose, a Kodak PixPro AZ255 digital camera was used. In this initial stage, a total of 228 images of heritage windows were collected.

In a second stage, additional images were incorporated from the ImageMG dataset, developed in a previous study by the present authors:
- https://doi.org/10.17632/nhfp5yd3jn.1
The ImageMG dataset comprises 6,449 images, of which 1,598 correspond to windows from historic churches located in the state of Minas Gerais. It is important to note that ImageMG contains five classes, with only a single class dedicated to windows. Therefore, it does not distinguish among different types of this key façade element.

In the data preprocessing and annotation stage, the most common types of windows in the photographed Brazilian historic churches were initially analyzed. Additionally, the Glossary of Architecture and Ornamentation of the Baroque period in Minas Gerais by Affonso Ávila et al. was consulted, which presents the most frequent window designs found in cultural heritage buildings in the state of Minas Gerais, particularly from the eighteenth and nineteenth centuries. Based on this analysis, four classes of heritage windows were defined for consideration in this study:

1) Balcony Door (Sacada): A full-height window that provides access to a balcony. It is typically located on upper floors and features a guardrail as a safety measure to prevent falls.

2) Sash Window (Guilhotina): a window consisting of two large panels that move vertically, allowing the opening or closing of the access. Another characteristic of this type is the use of glass panes.

3) Panelled Window (Almofada): a wooden window featuring prominent rectangular panels.

4) Plain Window (Calha): a simple wooden window without elaborate details, composed of continuous vertical elements spanning between its edges.

The class nomenclature was defined using both English and Portuguese terms (the latter in parentheses as a short form) to assist both international and local audiences in understanding the types of heritage windows evaluated.

Next, 629 photographs that did not correspond to any of the four analyzed window classes were removed.

The final dataset comprises 1,197 high-resolution images (4068 x 3456 pixels) and is divided into four classes, with the quantities detailed:

1) Balcony Door (Sacada): 657.
2) Sash Window (Guilhotina): 236.
3) Panelled Window (Almofada): 184.
4) Plain Window (Calha): 120.
Total: 1,197 images