Synthetic clear sky generation

Automated pipeline for generating synthetic clear-sky images from real sky images with clouds using the Chauvin et al. (2015) photometric model.

🚀 Usage

# Install dependencies
pip install -r requirements.txt

# Run on an image
python src/synthetic_clearsky_optimizer.py \
    --image path/to/image.png \
    --mask path/to/mask.png \
    --output output_directory

Input requirements:

• Image: RGB fisheye image (PNG, JPG, WEBP)
• Semantic mask: PNG with labeled regions:
  • RGB(66, 245, 84): Background
  • RGB(66, 135, 245): Clear sky
  • RGB(245, 66, 66): Cloud
  • RGB(245, 212, 66): Sun

Output files:

• best_synthetic_clearsky.png - Final synthetic clear-sky
• best_comparison.png - Side-by-side comparison
• optimization_results.json - Parameters and metrics

📊 Results

Demo 1: Arizona, february 2025

Demo 2: Copenhagen, july 2025

Demo 3: Paris, september 2025

Demo 4: Almería, june 2024

Demo 5: Copenhagen, september 2025

🎯 Method

4-stage hybrid optimizer:

1. Grid search - Tests multiple sun sizes and band configurations
2. Continuous refinement - Optimizes 21 clear-sky parameters (7 per RGB channel)
3. Sun enhancement - Adds realistic sun disk + bloom PSF
4. Sun parameter grid - Exhaustive search across 252 configurations

See TECHNICAL_DOCS.md for detailed algorithm description and mathematical models.

📄 License

See LICENSE file.

---

Version: Production
Last Updated: 2025-10-18
Status: ✅ Production Ready