Dataset: heterogeneous perception of urban space as a function of image attributes

The dataset contains results from an online survey where pairs of images depicting public places were presented to respondents. Respondents were asked to choose the image that best suited a randomly assigned qualitative attribute describing the place (safe, walkable, livable, beautiful, and wealthy). The survey was available in Spanish and English and also collected some sociodemographic characteristics of the respondent such as gender, age, nationality, current country of residence, educational level and main transportation mode.

Respondents were allowed to answer as many choice experiments as they wanted. The dataset contains 27311 choices from 1536 respondents. The dataset also contains attributes for all presented images, extracted with machine learning algorithms (semantic segmentation and object detection) as described by Rossetti et al. (2019) and Ramirez et al. (2020).

The dataset was used to estimate discrete choice models to forecast the probability of an image being labeled as safe, walkable, livable, beautiful, and wealthy. This model was then applied to approximately 120000 images of Santiago, Chile, obtaining an estimation of the “score” each place gets for each qualitative attribute. This data is also available in the dataset

Modeled perception of safety in the city of Santiago (left) and spatial distribution of per-capita income (right) according to the latest travel survey (EOD2012)

Since the data includes respondents characteristics, and the model accounts for individual heterogeneity, the scores were also calculated for males, females and different main transport modes.

Map of differences in perception of safety based on gender. Places with a larger difference in perception between genders are displayed in darker red and imply a lower perception of safety for women in that place

The dataset is free to use for non-profit and research. We ask to cite the following paper when using it:

Ramirez, T.,  Hurtubia, R., Lobel, H. and Rossetti, T. (2021). Measuring heterogeneous perception of urban space with massive data and machine learning: An application to safety. Landscape and Urban Planning,  208, 104002  https://doi.org/10.1016/j.landurbplan.2020.104002

However, if you use this dataset, it is probably a good idea to also cite the work by Rossetti et al. (2019), where a large part of the included data was generated.

Rossetti, T., Lobel, H., Rocco, V. and Hurtubia, R. (2019) Explaining subjective perceptions of public spaces as a function of the built environment: A massive data approach. Landscape and Urban Planning, 181, 169-178. https://doi.org/10.1016/j.landurbplan.2018.09.020

The dataset is available here. If you have questions please contact Hans Lobel (halobel[at]ing.puc.cl) or Ricardo Hurtubia (rhg[at]ing.puc.cl)

Example of detection of relevant objects (top) and semantic segmentation (bottom) of the images