🂡 CardVision AI

A deep learning-based Playing Card Classification application built from scratch using PyTorch and deployed with Streamlit.

The model classifies 53 different playing card classes (including Joker) from an uploaded image and displays the predicted card along with confidence scores.

---

🚀 Live Demo

🔗 Streamlit App: https://cardvision-ai.streamlit.app/

---

📸 Preview

✨ Features

• 🧠 Custom CNN built completely from scratch
• 🃏 Supports 53 Playing Card Classes
• 📷 Upload any playing card image
• 🎯 Predicts the card with confidence score
• 📊 Displays Top-5 Predictions
• ⚡ Dynamic Quantization (196 MB → 49 MB)
• 🌐 Interactive Streamlit Web App

---

🏗 Model Architecture

Input Image (224 × 224 × 3)

↓

Conv2D (3 → 32)
ReLU
MaxPool

↓

Conv2D (32 → 64)
ReLU
MaxPool

↓

Conv2D (64 → 128)
ReLU
MaxPool

↓

Flatten

↓

Linear (100352 → 512)

↓

ReLU

↓

Dropout (0.5)

↓

Linear (512 → 53)

↓

Prediction

---

📊 Training Details

Parameter	Value
Framework	PyTorch
Classes	53
Image Size	224 × 224
Optimizer	Adam
Loss Function	CrossEntropyLoss
Scheduler	ReduceLROnPlateau
Early Stopping	✅
Best Model Checkpoint	✅
Dynamic Quantization	✅

---

📈 Model Performance

• Training Accuracy: ~95%
• Validation Accuracy: ~84%

The model was additionally tested on unseen internet images and demonstrated good generalization under different lighting conditions.

---

📂 Project Structure

CardVision-AI/
│
├── app.py
├── best_model_quantized.pth
├── requirements.txt
├── README.md
└── images/
    ├── demo.png
    └── architecture.png

---

⚙ Installation

Clone the repository

git clone https://github.com/theghostloop/CardVision-AI.git

Move into the project directory

cd CardVision-AI

Install dependencies

pip install -r requirements.txt

Run the application

streamlit run app.py

---

🛠 Tech Stack

• Python
• PyTorch
• TorchVision
• Streamlit
• Pillow

---

🎯 Future Improvements

• Transfer Learning using ResNet/EfficientNet
• Grad-CAM Visualizations
• Webcam Live Prediction
• Batch Image Prediction
• Model Explainability

---

👨‍💻 Author

Prince

If you found this project useful, consider giving it a ⭐ on GitHub!