IITU Multi-Temporal Building Floor-Count Dataset: 12-Month Fixed-Viewpoint Construction Sequence

Published: 1 June 2026| Version 1 | DOI: 10.17632/ww7rt2rnk6.1
Contributor:
Arman Keresh

Description

This dataset provides a controlled, multi-temporal collection of street-level video recordings of a single reinforced-concrete building under active construction, captured from a fixed-viewpoint camera at the International Information Technology University (IITU) in Almaty, Kazakhstan. It was assembled to study the visual invariance of building floor-count estimation across construction stage, illumination, weather, and season, and accompanies the associated manuscript currently under peer review. The recordings span a twelve-month period, from May 2025 to April 2026. Because all footage depicts the same structure from the same vantage point, the building has a fixed ground-truth count of six structurally completed floors throughout the sequence; the collection is therefore intended as a robustness resource over environmental and construction-stage variation rather than as a source of floor-count class diversity. Between May and August 2025 the building was in a raw construction stage (exposed reinforced-concrete frame with no installed curtain-wall glazing or cladding); from September 2025 onward, progressive installation of glazing and exterior cladding transitioned it into a partial-finish stage. A completed stage is not represented, as facade installation had not finished by the end of the observation window. Capture protocol: Footage was recorded by a stationary street-level camera with a fixed field of view directed at the building facade. The recordings capture day-to-day and seasonal variation in illumination, weather, and construction progress over the full observation window.Intended use: This dataset supports research on construction-stage building monitoring, floor-count estimation from street-level imagery, and the robustness of visual models to seasonal, weather, and construction-stage appearance change. The annotated still-frame subset used in the associated manuscript will be added in a subsequent version of this dataset. License: CC BY 4.0 Keywords: floor-count estimation; construction monitoring; street-level imagery; reinforced concrete; multi-temporal; building stock; computer vision

Files

Steps to reproduce

This version of the dataset contains the raw video footage only. Recordings were captured by a stationary street-level camera with a fixed field of view directed at the building facade, at the International Information Technology University (Almaty, Kazakhstan), between May 2025 and April 2026. The footage documents continuous day-to-day and seasonal variation in illumination, weather, and construction progress over the full twelve-month observation window. Between May and August 2025 the building was in a raw construction stage (exposed reinforced-concrete frame with no installed curtain-wall glazing or cladding); from September 2025 onward, progressive installation of glazing and exterior cladding transitioned it into a partial-finish stage. The annotated still-frame subset used in the associated manuscript will be released in a subsequent version of this dataset. In that version, frames will be sampled from the footage at a minimum spacing of ten minutes (so that no two retained frames originate from the same short capture burst), manually verified (excluding frames with heavy occlusion or extreme low light), and annotated with a fixed floor-count label (6 structurally completed floors), a construction stage (Raw or Partial), a weather category, a time-of-day category, and the capture timestamp. A README file with full field definitions and value sets will accompany that version.

Categories

Computer Vision, Civil Engineering

Funders

  • Science Committee of the Ministry of Science and Higher Education of the Republic of Kazakhstan
    Grant ID: AP23488873

Licence