🥭 Mango 🥭

Mango is an open-source comprehensive trading card platform that combines computer vision analysis, metadata management, and collection tracking capabilities to provide a complete solution for card collectors and investors.

example - ArtiflectAI

System Architecture

The Mango platform consists of three main components:

mango-client

A sophisticated web application for collectors that provides:

• Custom collection creation and management
• Granular card filtering and organization
• Price tracking and analytics
• Portfolio value monitoring
• Integration with vision pipeline for card identification
• User-defined collection categories and tracking parameters

mango-metadata-api

Backend service built with Node.js that manages and serves trading card metadata. This service provides:

• RESTful API for card metadata queries
• Database management for card information
• Metadata validation and processing
• Integration with card pricing APIs
• User collection data management
• Price history tracking

mango-vision-pipeline

Containerized computer vision pipeline that combines multiple models for card analysis:

• MMDet-based object detection for detecting card objects in images
• Facebook's Segment Anything Model (SAM) for precise card segmentation
• Custom CNN for visual similarity search
• Docker-based deployment for easy scaling

mango-visual-search-training-eval-pipeline

Training and evaluation pipeline for the visual search component:

• Custom ResNet-based model training for card similarity
• Vector embedding generation for ground truth images
• Comprehensive model evaluation across different card types
• 3D visualization tools for embedding analysis
• Modular pipeline design for easy experimentation
• Automated evaluation across multiple test sets (normal, holo, full-art)

System Flow

1. Users can build collections through:
   • Manual card entry with granular details
   • Images processed by the vision pipeline
2. The vision pipeline processes uploaded images:
   • Object detection identifies and localizes cards
   • SAM provides precise segmentation
   • Visual search identifies similar cards
3. Metadata API enriches results with:
   • Detailed card information
   • Current market prices
   • Historical price data
4. Users can:
   • Track collection value
   • Monitor price changes
   • Create custom sets and track completion

Getting Started

Prerequisites

• Docker and Docker Compose
• Node.js 18+
• Python 3.8+
• MySQL
• CUDA-compatible GPU (recommended)

Installation

1. Clone the repositories:

git clone https://github.com/riley-livingston/mango-client
git clone https://github.com/riley-livingston/mango-metadata-api
git clone https://github.com/riley-livingston/mango-vision-pipeline
git clone https://github.com/riley-livingston/mango-vision-pipeline

2. Follow the individual setup instructions in each repository's README.

Contributing

We welcome contributions! Please see our Contributing Guide for details.

License

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

Contact

• GitHub: @riley-livingston
• Project Link: https://github.com/riley-livingston/mango

Acknowledgments

• Facebook Research for the SAM model
• MMDetection team
• Open source community