Related Experiment Video
Updated: May 5, 2026

Deep Neural Networks for Image-Based Dietary Assessment
Published on: March 13, 2021
Image dataset: Year-long hourly façade photos of a university building
Marc Roca-Musach1, Isabel Crespo Cabillo1, Helena Coch1
1Architecture, Energy and Environment Research Group (AiEM) - Universitat Politècnica de Catalunya, Spain.
Abstract:
This dataset presents 9596 photos of the East and West façades of a university building in Barcelona (Spain). Images were taken every hour for one year with two identical GoPro Hero 10 cameras. The façades are composed of a grid of 28×7 windows in the East and 28×6 windows in the West. Every window has a portion of fixed glass and an operable part for natural ventilation, and mobile solar protections (roller shutters) that users can control manually. These images are of special interest due to the lack of observation data on the user-building interaction with manually controlled adaptive façades. These data are a valuable source of information for interpreting and understanding the actual use of manually controlled adaptive façades, as well as developing usage models that could be implemented in energy simulations. Also, cropping all the windows from the full-façade photos result in 1.7 M images of individual windows, which can be easily used as a computer vision and machine learning exercise to read the position of the solar protections or the operable part of the windows.

