Swiss GlassScan: Estimating the Window-to-Wall Ratio of buildings for improving energy consumption models

Window-to-wall ratio (WWR) is a ghost variable in energy modeling: we know it's critical for thermal performance, yet we lack large-scale data for the Swiss building stock. 

In this challenge, you will transform Google Street View imagery into a large-scale architectural dataset. Using Google Street View API, you will:

• Extract & Segment: Use pre-trained segmentation models to identify facades and windows from street-level imagery.
• Rectify: Apply homographic transformations to correct perspective distortion, ensuring accurate area measurements.
• Quantify: Calculate the WWR and correlate it with metadata like building age, usage, and location.
• Predict: Train a regressor to estimate WWR for buildings where imagery isn't available, creating a comprehensive heat-demand map.

Heating accounts for a massive portion of Switzerland's CO2 footprint. Current energy models often rely on "average" assumptions about window sizes, leading to inaccurate renovation strategies. By automating the extraction of WWR, you are providing a valuable feature for precise thermal simulations. This tool enables policymakers and urban planners to identify inefficient districts and prioritize retrofitting where it's most needed, directly supporting the Swiss Energy Strategy 2050

• Metadata: Construction year, building category (residential, commercial),...
• API Access: Guidance on using the Google Street View Static API (10k free calls) to fetch facade images

• Python: Advanced
• Computer Vision: Intermediate
• Data Science: Experience with regression models (Scikit-learn, XGBoost,...)