Human Activity Units in Curated Architectural Images: Coded Data on Posture, Social Interaction, Age Group, and Use of Space
Description
This dataset records how people were pictured in a curated set of architectural images. It contains 218 Activity Units coded from 79 images that accompanied 48 project features on a single, widely read online architecture and design publication. An Activity Unit is a person or group engaged in one shared activity within an image, so one image can contain several units. Most units (208) came from features published in 2025. The remaining ten came from features published in 2021 and 2022. Each unit was coded for activity (intensity, rituality, implied duration), social and bodily relation (group size, interaction, posture, gaze, mobility), spatial context (location, proximity, objects used, spatial scale, mode of use), population (age group, gender, recorded accessibility cues, recorded functional adaptation), atmosphere, and frame attributes (type of imagery, implied time of day, weather cues). Tier 1 variables can be checked directly against an image. Tier 2 variables require interpretation and should be treated as descriptive. Two coders, both authors of the related publication, coded every unit jointly and entered a code only when both agreed. No independent double coding was carried out, so intercoder agreement statistics are not available. Accessibility cues and functional adaptation were recorded only when noticed, so a blank cell does not confirm absence. The release contains two files. HumanActivityUnits_Data_V1.xlsx is a Microsoft Excel workbook with three sheets: Data (218 rows, 42 columns, one row per Activity Unit), Codebook (variable list), and Notes. HumanActivityUnits_Codebook_V1.pdf documents the selection criteria, unit of analysis, coding procedure, every variable, the cleaning steps, and known limitations. Features and images carry anonymous codes (feature_id, image_id) that preserve the nesting of units within images and images within features. Source images, links, file names, the name of the publication, and coder identities were removed. The dataset can be used to reproduce the reported proportions, to apply the Visual Inclusion Audit to other image sets, or as a reference for studies of age, ability, and use of space in architectural representation. Findings describe one publication and should not be generalized to architectural imagery as a whole.
Files
Steps to reproduce
1. Images were taken from 48 project features on one online architecture and design publication. Every image had to show people in a plausible everyday activity. Photographs were also required to be candid, not posed or directed by the photographer or a stylist. Renderings were exempt from this rule, because their figures are always placed by the author. Full criteria are in the codebook. 2. In each image, every person or group engaged in one shared activity was coded as a separate Activity Unit, so one image can yield several rows. 3. Units were coded with a structured questionnaire (closed categories plus short phrases for activity, location, proximity, and objects). 4. Two coders viewed each image together and entered a code only when both agreed. Every code was a human interpretation; no computer vision or automated detection was used. Accessibility cues and functional adaptation were recorded only when noticed. 5. Derived fields were added: anonymous feature_id and image_id, feature_year, dominant posture, time and weather classes, age group flags, and the Aesthetic Habitation Script score (ahs_score5, 0 to 5; ahs_score4 without mobility). 6. Links, image files, file names, the publication name, and coder identities were removed. 7. To reproduce the results, compute unit level proportions over the 218 rows with 95% Wilson score intervals, count features with at least one unit per category by feature_id, and count frame attributes once per image_id (79 images). 8. Units are nested in images and features, so no significance tests were run. Account for this nesting, for example by clustering on feature_id, before testing associations.
Institutions
- Binus UniversityJakarta, Jakarta