A Structured Repository of Architectural Visualisation Metadata, Visual Characteristics, and Machine-Vision Descriptions
Description
This dataset consists of a comprehensive structured spreadsheet containing 296 entries that document the intersection of architectural visualisation, sustainability claims, and artificial intelligence interpretation. The data was compiled through a rigorous content analysis of secondary sources from leading digital architectural platforms such as ArchDaily and Architizer. It serves as a multidimensional record that bridges the gap between technical visual metadata and qualitative design narratives. The repository is organised into several key thematic clusters. The first section contains essential project metadata, including project titles, architectural firms, geographic locations, and scales ranging from individual building components to urban districts. This allows researchers to categorise the data based on global trends and architectural typologies. The second cluster focuses on the characteristics of the visual media itself. It records static image, the specific visualisation type such as 3D renders or photography, and the artistic point of view. Furthermore, it incorporates detailed perceptual variables including brightness, tonal range, colour temperature, saturation, and composition. These metrics provide a quantitative basis for studying the atmospheric and technical qualities of architectural imagery. A significant portion of the dataset is dedicated to sustainability. It captures the specific claims made by architects and developers regarding environmental performance. This is further broken down into sustainable design attributes, such as passive cooling or renewable energy systems, and the specific sustainable materials highlighted in the visuals. By categorising these elements, the data enables an investigation into how environmental responsibility is communicated visually to the public. A unique feature of this dataset is the integration of AI-based visual inspection results. Each entry includes descriptive tags and adjectives derived from AI image-to-text processes. This allows for a direct comparison between human-authored project descriptions and the objective interpretations provided by computer vision models. The dataset is provided as a single Excel file to maintain the integrity of the cross-referenced variables. Manual untuk menggunakan dataset ini tertulis dalam A Structured Repository of Architectural Visualisation Annotation Manual (PDF). Users should note that while the metadata is exhaustive, the images themselves are referenced via permalinks to respect original copyright holders. Meanwhile, high resolution charts for some key features are in the Distribution Charts folder. This collection represents a significant contribution to the digitisation of architectural research and the study of human-centric sustainable environments.
Files
Steps to reproduce
Four researchers collected and coded the dataset. Weekly meet for two months to pilot the codebook, align operational definitions and triangulate interpretations. 1. Preparation and sampling • Records were sourced from digital architectural platforms and retained original metadata and permalinks. 2. Codebook and training • A codebook with definitions, examples and inclusion rules was iteratively refined. • Coders piloted several of records to resolve ambiguities and standardise rules. • Codebook definitions are embedded in the Microsoft Forms, for easier data entry. 3. Feature-level coding • Project metadata: transcribed verbatim and standardised. • Media metadata: classified by type, viewpoint and visualisation technique. • Sustainability claims and materials: extracted verbatim and mapped to controlled categories. • Adjectives and semantic descriptors: lemmatised and coded for theme and valence; AI tags recorded verbatim. • Perceptual variables (brightness, tonal range, colour temperature, contrast, saturation, sharpness, hue, texture, composition, depth): interpreted on ordinal scales using visual anchors; objective image metrics used where available. • Binary and categorical fields: coded with explicit rules for missing or ambiguous entries. 4. Triangulation and quality control • Each item was coded independently by at least two researchers. Disagreements were discussed and resolved in weekly meetings by consensus or adjudication. • Consistency of assessment was maintained through ongoing progress monitoring and weekly review discussions, rather than by means of formal statistical measurement. 5. Finalisation • Codes were standardised, missing-value rules applied, and permalinks validated. Raw text fields were preserved alongside coded variables to support reproducibility.
Institutions
- Binus UniversityJakarta, Jakarta