MTGScan Cards is an open-source project for detecting, identifying, and analyzing physical Magic: The Gathering cards using computer vision and machine learning.

This organization hosts the fullstack application, keypoint detection model, and feature extraction workflows used in the MTGScan pipeline.

🖥️ Fullstack web application for scanning and identifying MTG cards

• Built with React (frontend) and Flask (backend)
• Integrates OpenCV + TensorFlow.js for real-time card corner detection
• Supports scanning via desktop camera or mobile QR offloading
• PostgreSQL-based collection tracking and card database
• Dockerized for local or server deployment

📐 ML model for card quadrilateral corner prediction

• Heatmap-based keypoint regression using TensorFlow/Keras
• Trained on a synthetic dataset of MTG card images with randomized distortions
• Uses MobileNet or ResNet backbones with SoftArgmax decoding
• Outputs 4-point quadrilateral coordinates per frame
• Exportable to TensorFlow.js for WebGL inference w/ float16 quantization

🧠 Feature extraction pipeline for card identification

• Uses OpenCV + SIFT/RootSIFT for feature vector generation
• Batch-processing pipeline that stores descriptors in HDF5
• Builds FAISS indices for fast visual similarity search