🐼 Jordan-Panda-AI: End-to-End Sneaker Classifier

Jordan-Panda-AI is a professional-grade Data Engineering and Computer Vision project designed to automate the detection of Nike Jordan 1 "Panda" sneakers in secondary markets. This project serves as a comprehensive proof-of-concept for high-scale data acquisition and deep learning inference.

🚀 Overview

The system automates the entire machine learning lifecycle:

1. Data Ingestion: Automated scraping of Nike (Official) and Wallapop (Market) using anti-bot bypass techniques.
2. Preprocessing: Image augmentation to diversify training samples and prevent early-stage overfitting.
3. Model Training: A custom Convolutional Neural Network (CNN) built with PyTorch, optimized through iterative evaluation.
4. Inference: A production-ready script to classify local images with confidence scoring.

🛠️ Tech Stack

• Automation: Selenium & undetected-chromedriver for advanced web scraping.
• AI/ML: PyTorch, Torchvision (Transforms), and PIL.
• Data Handling: Pandas for metadata management and Pathlib for robust file system navigation.
• Environment: Python 3.10+.

📦 Installation & Setup

1. Clone the repository:

   git clone [https://github.com/YOUR_USERNAME/Jordan-Panda-AI.git](https://github.com/YOUR_USERNAME/Jordan-Panda-AI.git)
   cd Jordan-Panda-AI

2. Create a virtual environment:

   python -m venv venv
    # Windows:
    venv\Scripts\activate
    # Linux/Mac:
    source venv/bin/activate

3. Install dependencies:

   pip install torch torchvision pillow selenium undetected-chromedriver webdriver-manager pandas requests
   

🎯 Educational Objectives & Results

This project was developed as part of a Data Engineering specialization, focusing on mastering data pipelines and real-world AI challenges.

• Final Accuracy: 80.26% on the validation set after 10 epochs.
• Optimization: Identified a significant improvement (from 57% to 80%) by balancing real-world market data with high-quality catalog images.
• Lessons Learned: Successfully navigated the "Overfitting" phase, identifying that extremely low training loss ($0.0091$) requires a more diverse negative-sample dataset to maintain high real-world precision.

⚠️ Industrial Scalability

To transition this PoC into a 99% accuracy commercial tool, the following enhancements are required:

1. Dataset Expansion: Scaling from hundreds to thousands of unique samples.
2. Transfer Learning: Implementing pre-trained architectures like ResNet50 or EfficientNet.
3. Hardware Acceleration: Training on high-performance GPU clusters to support deeper architectures.

---

Author: raess1593 – Data & AI Engineer Student