This project is an experimental implementation of Named Entity Recognition (NER) from images using the Phi 3.5 Vision Instruct model. The system takes an image and extracts meaningful entities directly from it. It was developed as part of a hackathon challenge, and as such, it remains a work in progress. Several features and optimizations are planned but have not yet been completed.
This project was built for a hackathon and is not a complete, production-ready solution. The current implementation works in basic scenarios but lacks extensive error handling, performance optimizations, and support for all edge cases.
- Uses Phi 3.5 Vision Instruct model for extracting entities from images.
- Input: An image containing text (e.g., receipts, documents, posters).
- Output: Named entities such as person names, organizations, and dates.
- Limited preprocessing for handling noise in image data.
- Hackathon build — expect incomplete features and limitations.
- Python 3.8+
- PyTorch (CUDA support recommended for faster processing)
- Required libraries:
transformerstorchvisionPillow
🎊🎊 top 10% 🎊🎊
