129 AEC Datasets, Verified

10 million automatically reconstructed 3D buildings covering the entire Netherlands in LoD1.2, LoD1.3, and LoD2.2.

Test datasets for 3D file formats focusing on IFC, OBJ, FBX, GLTF, XKT, 3DS, MAX and other 3D formats for benchmark comparisons.

325+ million points from 47 Nagara-style temple structures. India's first benchmark dataset for temple architecture semantic segmentation.

Full text of 2010 ADA Standards for Accessible Design and ADA/ABA Accessibility Guidelines from the U.S. Department of Justice.

4,800 questions across 23 task types from ECADI + Tongji University covering building codes, design calculations, and construction management.

Benchmark for evaluating AI ability to process architectural floor plans. Text extraction ~95%, spatial reasoning 55-75%, symbol counting 39-51%.

300+ industry-standard A-Series, B-Series, C/D/E/G-Series contract forms including A101, A201, B101. Most widely used US construction contracts.

1,100+ manufacturers, 33,000+ building products organized by CSI MasterFormat with downloadable specs, AutoCAD, and Revit BIM families.

17 annotated heritage scenes (millions of labeled 3D points) plus 10 unlabeled scenes. 10 semantic classes, UNESCO World Heritage sites included.

Semantically-enriched photo-realistic 3D architectural models for semantic segmentation of building exteriors. 4 building types from Hong Kong.

413,062 chunks from 5,538 CAD drawings across 27 categories. Largest architectural CAD dataset for panoptic symbol spotting. NeurIPS 2025.

AI training dataset for real estate/architecture with open-source AGPL-licensed floor plan to IFC annotation pipeline. Presented at ICCV Paris 2024.

13,000+ architect salary surveys across the United States with data visualization. Covers firm size, location, and career level.

Architectural heritage elements for classification at 64×64 resolution, organized by element class in folder-per-class structure.

10,113 images across 25 architectural styles from multiple time periods and mixed sources. Organized in 25 class folders.

44,000 BIM element samples (4,000 per class × 11 classes). First large-scale BIM element classification dataset from Seoul National University of Science and Technology.

Curated directory of 210+ million semantic 3D buildings from 21 countries and 65+ cities including Netherlands, Poland, Luxembourg, Germany, and Singapore.

Curated list of US federal procurement data resources, APIs, and utilities including pysam, SamDotNet, and procurement-tools Python wrappers.

Python tool that extracts geometric data (CSV/WKT) from IFC models and generates 2D floor plans (PNG) for large-scale dataset creation.

3,965 annotated RGB images from 50+ buildings (10–60 years old) covering 6 defect types: algae, major/minor crack, peeling, spalling, stain.

Open standard schema enabling sub-second data export from Revit models to DuckDB. Eliminates traditional BIM data bottlenecks.

116.5M points from 25 real-world scans, 382 rooms, 8,710 m². First openBIM-based scan-to-BIM dataset with IFC ground-truth models.

ML competition ($20K AUD) for automating building data classification using Brick schema. Based on NeurIPS 2024 Building Timeseries Dataset.

Largest annotated floor plan dataset for Document Analysis. Aggregates ROBIN, SESYD, and web sources with bounding boxes and paragraph descriptions.

598 images with 10 classes (balcony, car, facade, fence, shop, street, traffic, vegetation, window) in COCO JSON, Pascal VOC, and YOLOv8 formats.

53.6M measurements from 3,053 meters in 1,636 buildings over 2 years, plus 900K synthetic profiles from NREL. NeurIPS 2023.

71,334 automated architectural plans in DrawScript vector format generated using shape grammar and visual programming. AI training dataset.

Framework for processing large-scale OpenStreetMap urban data in Grasshopper/Rhino, enabling parametric urban modeling workflows.

Research data from Center for the Built Environment comparing WELL vs LEED building IEQ satisfaction scores. Free download.

Building permits from major US/global cities via open data portals: NYC, Chicago, Seattle, Los Angeles, San Francisco, Vancouver, Melbourne.

606 rectified facade images with manual pixel-level annotations across diverse architectural styles from the Czech Technical University.

862 sentences, 4,297 entities, 4,329 relations from 33 regulatory documents (1,595 pages) annotated by 12 experts. England and Finland.

Tabular construction cost and estimation records for training AI-driven cost estimation and bid analysis models.

Real construction environment PPE compliance/non-compliance images in Ultralytics YOLO format for worker safety monitoring.

10,013 construction site images (7,009 train / 3,004 test) for testing vision-language models on construction inspection tasks.

Smartwatch data on residents' perceptions of urban noise and heat collected via the Cozie smartwatch platform. Kaggle ML competition.

Steel bridge images with pixel-wise fatigue crack annotations from Rijkswaterstaat and ProRail, covering cracks, corrosion, and defect-free structures.

5,000 real-world floor plan images from Finnish real estate with 80+ annotation categories including rooms, doors, windows, walls, and stairs.

Building permits, zoning GIS data, development permits, property maintenance codes, land development codes, and state energy codes from the US government.

150,000+ verified Environmental Product Declarations (EPDs). Largest open-access database for embodied carbon in building materials.

All US federal regulations including HUD, ADA, OSHA, and fire codes in XML bulk format with continuous updates.

104 rectified Hausmannian building images from Paris with 7 semantic classes (wall, window, sky, shop, balcony, door, roof, chimney).

60 annotated non-rectified facade images with 4-class and 8-class annotation variants covering irregular patterns.

15,663 real-world CAD floor plans with fine-grained annotations for 30 object categories (doors, windows, furniture, equipment). ICCV 2021.

Performance-optimized binary BIM format rendering millions of objects with LOD. Reduces 2GB IFC files to ~200MB with geometry deduplication.

Basic construction contract templates (lump sum, cost-plus, time & materials, design-build) from TemplateLab, Mastt, Jotform, and eForms.

72 million contracts from 42 countries (2006–2021) with buyer/supplier info, geolocation, product classification, price, and corruption risk indicators.

2.75 billion 3D building structures at 3-meter resolution derived from ~800,000 satellite images. Global coverage of footprints, heights, and LoD1 models.

LLM + Knowledge Graph pipeline for smart building management using IFC data parsed into Neo4j graphs with ifcopenshell.

11,123 high-resolution bridge images with 6 defect types (cracks, spalling, seepage, honeycomb surface, exposed rebar, holes). Nature 2025.

Defect detection dataset for heritage buildings with bounding box and segmentation mask annotations for inspection applications.

Photogrammetric images and annotated point clouds for 5 heritage buildings following ArCH classification standards.

Building permit data for all US Metropolitan Areas, Central Cities, and Suburbs from the U.S. Dept. of Housing and Urban Development.

Synthetic fault data for building fault detection using EnergyPlus simulation + GANs. 50% improvement in FID scores, classifier accuracy 0.82 → 0.94.

Nature Communications framework for automated floor plan generation using hypergraphs. Open-source Python code with graph data structures.

624 high-resolution buildings with 43,277 annotated windows (avg 4056×3856 px) from Hangzhou, China. Includes homography matrix annotations.

Full text of IBC 2024/2021, IRC, IFC (Fire), IMC, IPC, IECC and more. Searchable online free; REST API via Code Connect (paid).

Node-based visual editor for Information Delivery Specifications (OpenBIM). Drag-and-drop IDS rule creation without XML coding.

13,485 question-answer pairs covering BIM and IFC domain knowledge for training/testing LLMs on BIM-specific queries.

ML pipelines for automated 3D IFC object classification achieving ~98% accuracy using GNN, Stacked RF, and Deep Learning.

Browser-based IFC validation using IDS standard and IfcOpenShell. Validates IFC models against Information Delivery Specifications.

21 IFC model projects with 1,027 QA pairs for testing LLMs on natural language queries to IFC building information retrieval.

Converts IFC BIM files to DuckDB for high-performance sub-second SQL queries on building data. Open-source hackathon project.

Free, open-source browser-based embodied carbon calculator. Converts IFC models to CO2-equivalent data using Swiss national carbon databases.

~19,000 CAD models across 65 IFC classes (IFCNetCore: 7,930 objects, 20 classes) extracted from ~1,000 IFC models for BIM entity classification.

National Planning Application Database with spatial and tabular data. Dublin City Council applications from 2003 to present.

1,057 high-quality facade images from 104 countries (1895–2023) with 6 segmentation classes: Background, Plant, Wall, Window, Door, Fence.

800+ Dutch apartment IFC models from real housing projects with annotated floor plans (WKT), material data, and spatial graphs.

Highly irregular and diverse building facade images from the LabelMe segmentation dataset, challenging standard segmentation approaches.

Open standard for exchanging LCA results and EPD data across software platforms, born from Denmark's 2023 building regulation on LCA submissions.

Learning resource and implementation guide for architects building RAG (Retrieval-Augmented Generation) systems with vector databases.

9m lattice mast structure with 18 reversible damage cases across 6 levels. Accelerometers, strain gauges, and temperature sensors for SHM research.

14,471 UAV-collected building images across 6 structure types (steel, concrete, wood, brick) with defects including cracks, leakage, and corrosion.

1.8M images (Standard) to 8M (Challenge) across 365 scene categories including many building-related categories. Download available on Kaggle and official.

954 high-resolution multi-unit residential floor plans from 165 Chilean projects with wall and slab polygon annotations.

~100,000 modern architecture building photographs on Kaggle. No labels or annotations — raw photography dataset.

~16,800 annotated apartment-level floor plans with rich graph annotations covering windows, doors, orientation, and 22 room subtypes.

Benchmark dataset for floor plan generation of building complexes, published at ECCV 2024. Available on Kaggle, GitHub, and arXiv.

Standardized zoning data covering 200+ regulatory characteristics across the US. Most comprehensive national zoning dataset.

Full text of NFPA 101 (Life Safety), NFPA 72 (Fire Alarm), NFPA 13 (Sprinkler), and 300+ fire codes via NFPA LiNK.

100+ country procurement datasets in standardized Open Contracting Data Standard covering planning, tender, award, contract, and implementation stages.

~125 million 3D building geometries across the entire US derived from USBuildingFootprints, available on AWS S3.

Model Context Protocol server connecting AI assistants to OpenBIM data. Enables AI to query, reason, and answer questions about IFC models.

Framework converting OpenBIM IFC models to Finite Element Method analysis meshes. Python-based IFC to FEM pipeline.

Systematic catalog of 51+ open-access visual datasets for construction AI covering safety, quality, progress, equipment across multiple modalities.

Curated collection of high-quality BIM files in IFC2X3, IFC4, and IFC4x3 covering architectural, structural, and MEP models.

400 facade photographs paired with hand-labeled semantic maps (walls, windows, doors). The classic pix2pix training dataset.

Live planning application and development proposal data from all London boroughs, updated daily by the Greater London Authority.

National planning and housing data from all English Local Planning Authorities with map view, search, and download. CSV, JSON, GeoJSON.

Part-of-speech tagged building code corpus used to train Bi-LSTM RNN with BERT for building code NLP. 95.11% precision POS tagger.

12+ Python utilities for IFC file manipulation and BIM workflow automation. Open-source scripts for common IFC operations.

Browser-based IFC analysis for construction quantity takeoff. No uploads or installation required. Extracts quantities directly from IFC for cost estimation.

1,000+ paired samples each containing a parametric 3D Revit BIM model and annotated 2D floor plan, for BIM automation research.

17,000 vector-graph floor plans with precise wall, door, window, and functional space annotations. Graph representations with NetworkX compatibility.

Government, for-profit, and non-profit RFPs with a dedicated architecture category. Free listings with no subscription required.

510 black-and-white architectural floor plans plus 122 scanned documents (ROBIN++) for document analysis and automatic retrieval.

Deep learning model reconstructing 3D roof models from high-resolution aerial/satellite imagery for automatic LoD2 building generation.

80,788 densely annotated floor plans from real Asian residential buildings with room types, wall distinctions, and door types. Used by HouseGAN.

Architecture image dataset (2022) on HuggingFace with Parquet metadata format.

6 areas, 3 buildings, ~270 rooms with 13 semantic categories per point (ceiling, floor, wall, beam, column, window, door, furniture).

All US federal government contract opportunities (RFPs, solicitations, awards) for all agencies. REST API, CSV bulk downloads, PostgreSQL snapshots.

56,000+ images (256×256 px) of cracked and non-cracked concrete from bridge decks, walls, and pavements. Crack widths 0.06mm–25mm.

1,000 synthetic floor plans (16 architectural symbol models). Extended SFPI version: 10,000 images with ~300,000 furniture items.

200,000+ permits (5 years) to 1.1M+ records (1980-2019) with estimated cost, description, permit type, location. Weekly updates from DataSF.

180M+ AI-enriched permits across 30M US addresses with inspection pass rates, contractor profiles, and 85% US population coverage.

20,000+ images with 15 object classes covering workers, materials, machines, and layouts from multiple construction sites across conditions.

Novel 3D large language model for processing point cloud data with architectural spatial understanding and natural language output.

Demographically balanced global urban visual perception dataset from a 1,000-person survey across multiple cities. Published in Nature Cities.

Python library for structural engineering calculations following Eurocode and fib Model Code standards.

6.2 million synthetic residential buildings at LOD-4 with 3D wireframe graphs, floor plan images, and LiDAR-like roof point clouds.

2,500 synthetic single-family floor plans from T0 (2 rooms) to T4 (10 rooms) typologies, in black-and-white and color-coded versions.

926 annotated images (68.5 GB) from 6 UAV flights with 5 channels (RGB + thermographic + height). 6,927 thermal bridge annotations.

ALL EU public procurement notices since 1993 — the world's largest procurement dataset. REST API, XML bulk downloads, CSV subsets.

High-resolution building and unique architecture images on HuggingFace with Parquet metadata. Full-resolution architectural photography.

Open-source 3D topological modeling library integrated with Fragments for BIM. Python-based 3D topology data with IFC integration.

33 annotated building facades with ~333 million annotated LiDAR points. Semantic segmentation: windows, doors, balconies, moldings.

21GB of UAV RGB and thermal imagery for 3D building reconstruction and thermal anomaly detection, including 3D meshes.

UK government procurement opportunities and awarded contracts in OCDS format. API and CSV bulk downloads via data.gov.uk.

Unified classification system for UK construction covering products, systems, activities, and spaces. Used for BIM and specification classification.

80+ US jurisdictions, 190K+ local amendments, 6M+ code sections with AI Copilot for code research.

All US federal spending on contracts, grants, and awards back to FY2001. Filterable by NAICS codes (23xxxx for construction). REST API and CSV.

Open-source Python package for automated 3D city model generation from OpenStreetMap. Outputs voxel grids, GeoJSON, building heights, and vegetation data.

9,485 church images with 631 bounding box annotations and architectural style labels from Wikipedia. NeurIPS 2021 Datasets & Benchmarks.

Contract awards for World Bank-funded IDA/IBRD projects worldwide with Global Public Procurement Database on country procurement systems.

5,598 satellite image pairs (1024×1024) across 11 disaster events with 4-level damage labels covering hurricanes, wildfires, floods, and earthquakes.

Construction site monitoring combining YOLOv11 object detection with Autodesk ACC for automated progress tracking and safety monitoring.

1 year of continuous accelerometer monitoring from Z24 highway bridge with progressive controlled damage. The most popular SHM benchmark.

601 million annotated points in 5 and 15 class variants for large-scale facade semantic segmentation. Published at WACV 2025.

71,474 panoramas from 1,524 real unfurnished US homes with annotated 2D/3D floor plans from 20 US cities. CVPR 2021.