vis2attr - Vision Language Model for Attribute Extraction

Turn item photos into structured attributes (brand, colors, materials, condition) using Visual Language Models. Output strict JSON with per-field confidence scores.

High-Level Architecture

flowchart LR
    A[📁 Images<br/>JPG/PNG/WebP] --> B[📝 Prompt<br/>Jinja2]
    B --> C[🤖 VLM<br/>Mistral/OpenAI]
    C --> D[📊 Parse<br/>JSON]
    D --> E[⚖️ Decide<br/>Thresholds]
    E --> F[💾 Store<br/>Parquet]
    F --> G[📈 Export<br/>CLI]
    
    CONFIG[⚙️ Config<br/>YAML] --> B
    CONFIG --> C

Loading

🚀 Quick Start

# Install
uv venv && source .venv/bin/activate
uv pip install -e .

# Set up API key
export MISTRAL_API_KEY=your_api_key_here

# Run analysis
vis2attr analyze --input ./images --output ./predictions.parquet

# Batch processing with custom config
vis2attr analyze --input ./items --batch --config ./my-config.yaml --verbose

# Generate reports
vis2attr report --predictions ./predictions.parquet --format summary

📊 Status

Core infrastructure complete with working pipeline. Ready for production improvements.

Status	Component	Description
✅	Core Pipeline	Complete data models, configuration, Mistral provider, JSON parser
✅	File Processing	Image ingestor with EXIF stripping, comprehensive test suite
✅	CLI Interface	Full analyze command with batch processing, verbose logging, and result export
✅	Storage System	Parquet-based storage with lineage tracking and metadata storage
🚧	Decision Rules	Simple threshold-based decisions, needs sophisticated quality gates
🚧	Report Generation	CLI command structure ready, needs full implementation
❌	Metrics & Logging	Basic logging only, needs structured metrics collection
❌	Additional Providers	Only Mistral implemented (OpenAI, Google, Anthropic planned)

📚 Documentation

• Architecture - System design, data flow, and component overview
• CLI Reference - Complete command-line interface documentation
• Configuration - Configuration options and schema definitions
• API Reference - Data models, schemas, and programming interface
• Contributing - Development guidelines and contribution process

🎯 Design Principles

• Schema-first & config-driven: No hard-coded fields
• Ports & adapters: Swappable implementations via factory patterns
• Type safety: Comprehensive data models with validation
• Testability: Full test coverage for all components

---

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.