# Archilyse Research - Full Context

> Archilyse is a research project that transforms floor plan archives into simulation-dense context graphs. It provides verified building data, open datasets, and an AGPL-licensed pipeline for spatial analytics and AI applications in the built environment.

## Core Concepts

### Dense Context Graphs
Rooms are embedded in full spatial hierarchy (room → unit → floor → building → surroundings) with 367 simulation columns per area. This enables AI models to understand spatial relationships and performance characteristics simultaneously.

### Simulation Densification
Physically grounded signals are computed for every room:
- **Viewshed Analysis**: What is visible from each point (sky, greenery, buildings, water, mountains, streets, railway)
- **Daylight Simulation**: Sun position and direct/indirect light throughout the day and seasons
- **Spatial Connectivity**: Graph-based metrics including closeness centrality, betweenness centrality, and distances to key rooms
- **Traffic Noise**: Acoustic exposure from nearby roads and railways

### Verified Data
- Median area deviation <1.2% against portfolio indices
- Mandatory QA gates at each pipeline stage
- Human-in-the-loop verification for geometry and classifications

---

## Datasets

### Swiss Dwellings
- **URL**: https://archilyse.standfest.science/swiss-dwellings
- **Download**: https://zenodo.org/records/7788422
- **DOI**: 10.5281/zenodo.7788422
- **License**: CC BY 4.0 (commercial use allowed)
- **Scale**: 45,000+ apartments, 3,100+ buildings, 370,000+ rooms
- **Features**: 367 simulation columns per area
- **Coverage**: Swiss residential buildings

**Schema Overview**:
- `area_id`: Unique identifier for each room
- `apartment_id`: Links rooms to apartments
- `building_id`: Links apartments to buildings
- `area_type`: Room classification (living, bedroom, kitchen, bathroom, etc.)
- `area_size`: Verified floor area in m²
- Simulation columns: `view_*`, `sun_*`, `connectivity_*`, `noise_*`

**Citation**:
```bibtex
@article{standfest2022swiss,
  title={Swiss Dwellings: A large dataset of apartment models including aggregated geolocation-based simulation results covering viewshed, sun, and noise},
  author={Standfest, Matthias and Wiesmann, David and Staudacher, Ursina and Cozza, Stefano and Hagenauer, Sebastian and Jovanovic, Viktor},
  journal={Scientific Data},
  volume={9},
  pages={306},
  year={2022},
  publisher={Nature Publishing Group}
}
```

### Modified Swiss Dwellings (MSD)
- **URL**: https://archilyse.standfest.science/modified-swiss-dwellings
- **GitHub**: https://github.com/caspervanengelenburg/msd
- **Publication**: ECCV 2024 Workshop on Computational Aspects of Deep Learning
- **Scale**: 5,372 floor plans, 18.9K units
- **Purpose**: Graph benchmark for floor plan auto-completion and generation

**Benchmark Task**: Given a functional diagram (graph_in) and building structure (struct_in), predict the full layout (full_out).

**Derivation**: Curated from Swiss Dwellings v3.0.0 with:
- Feature removal for privacy
- Residential-only filtering
- Near-duplicate removal
- Medium-to-large scale filtering (≥15 areas, ≥2 Zone 2 areas)

**Citation**:
```bibtex
@inproceedings{vanEngelenburg2024msd,
  title={Modified Swiss Dwellings: A Machine Learning Benchmark for Floor Plan Auto-Completion},
  author={van Engelenburg, Casper and Jimeno Yepes, Antonio and Nardelli, Pietro},
  booktitle={ECCV 2024 Workshop on Computational Aspects of Deep Learning},
  year={2024}
}
```

---

## Pipeline

- **URL**: https://archilyse.standfest.science/pipeline
- **GitHub**: https://github.com/Archilyse/Archilyse
- **License**: AGPL (open source, copyleft)
- **Throughput**: 500-800 m²/h

### Six Pipeline Stages

1. **Floorplan Upload**: Intake of scanned or digital floor plans
2. **Georeferencing**: Align to real-world coordinates and cadastral data
3. **Annotation**: Web-based editor for wall, door, window, and feature placement
4. **Classification**: Room type assignment with ML-assisted suggestions
5. **Linking**: Connect units across floors and associate with portfolio indices
6. **Enrichment**: Run simulations (view, sun, noise, connectivity) and export

### Output Formats
- Tabular data (CSV, Parquet)
- IFC models (LOD 100-200)
- GeoJSON for spatial queries
- Graph representations for ML

---

## Industry Applications

- **Property Valuation (AVMs)**: Performance features reduce prediction error by ~50%
- **ESG Reporting**: Energy reference areas and construction quantities
- **Facility Management**: FM-ready spatial databases and IFC exports

### Data Refinement Levels
1. Annotated 2D models
2. Verified digital twins linked to client indices
3. Dimensions + simulation signals
4. Portfolio benchmarks
5. Predictive models (e.g., AVMs)

---

## Quick Start

**Google Colab Notebook**: https://colab.research.google.com/drive/1ln7y04SJ3xawyOK_ayeY3IQ3PGecLWnt

---

## Contact

- **Website**: https://archilyse.standfest.science/contact
- **Topics**: Research collaborations, commercial licensing, data access inquiries

---

## Additional Publications

### Bias Reduction (Foundational)
- **Title**: Reducing Bias for Evidence-Based Decision Making in Design
- **Author**: Matthias Standfest
- **Venue**: Design Computation Input/Output 2022
- **Publisher**: Springer
- **URL**: https://www.springerprofessional.de/en/reducing-bias-for-evidence-based-decision-making-in-design/23505652

This paper establishes the theoretical foundation for the project's bias-reduction paradigm, proposing a data refinery approach to reduce cognitive bias in design decisions.

```bibtex
@incollection{standfest2022bias,
  author    = {Standfest, Matthias},
  title     = {Reducing Bias for Evidence-Based Decision Making in Design},
  booktitle = {Design Computation Input/Output 2022},
  year      = {2022},
  publisher = {Springer},
  url       = {https://www.springerprofessional.de/en/reducing-bias-for-evidence-based-decision-making-in-design/23505652}
}
```

### PhD Thesis (2017)
- **Title**: Applying Emergent Features of Architectural Geometry
- **Author**: Matthias Standfest
- **Institution**: ETH Zurich
- **Year**: 2017
- **DOI**: 10.3929/ethz-b-000161416

---

## Licensing Summary

| Asset | License | Commercial Use |
|-------|---------|----------------|
| Swiss Dwellings dataset | CC BY 4.0 | ✅ Yes |
| MSD benchmark | CC BY 4.0 | ✅ Yes |
| Archilyse pipeline | AGPL | ⚠️ Copyleft (contact for commercial) |