photo_judgers/docs/scripts.md

Project Scripts

Setup Scripts

setup.bat (Windows) / setup.sh (Linux/Mac)

Creates a new conda environment and installs all required packages from requirements.txt.

What it does:

1. Checks that conda is installed
2. Creates a conda environment named photo_judgers with Python 3.11
3. Installs all packages from requirements.txt:
   • torch, torchvision, torchaudio (PyTorch with CUDA 12.1)
   • transformers (HuggingFace)
   • simple-aesthetics-predictor (LAION aesthetic predictor)
   • Pillow (image processing)
   • tqdm (progress bars)
   • numpy (numerical operations)

Usage:

# Windows
setup.bat

# Linux/Mac
chmod +x setup.sh
./setup.sh

Run Scripts

run.bat (Windows) / run.sh (Linux/Mac)

Activates the conda environment and runs the scorer.

What it does:

1. Activates the photo_judgers conda environment
2. Runs python -m src.scorer with any additional arguments passed through

Usage:

# Windows
run.bat

# Linux/Mac
chmod +x run.sh
./run.sh

Optional arguments:

• --input <path> — Directory or file to score
• --output <path> — Output JSON file (default: output.json)
• --threshold <value> — Score threshold for keep/reject (default: TBD)

Project Structure

photo_judgers/
├── src/
│   ├── __init__.py
│   ├── scorer.py          # Menu + proxy (entry point)
│   └── scorers/
│       ├── __init__.py
│       ├── base.py         # BaseScorer abstract class
│       └── laion.py        # LAION V2 implementation
├── requirements.txt
├── setup.bat / setup.sh
├── run.bat / run.sh
├── output.json             # Generated by scorer
├── docs/
├── test/
└── .pi/

Adding a New Scorer

1. Create src/scorers/<name>.py with a class that follows the LAION pattern
2. Register it in src/scorer.py by adding to the SCORERS list
3. Add any new dependencies to requirements.txt
4. Update src/scorers/__init__.py to export the new class