CIFAR-10 RNN Image Classifier
An end-to-end deep learning project for classifying CIFAR-10 images using a Recurrent Neural Network (LSTM) built with PyTorch. Includes a modern web interface for real-time image classification.
🌟 Features
- Custom RNN Architecture: Bidirectional LSTM layers with dropout
- Complete Training Pipeline: Automated training with validation, checkpointing, and visualization
- Comprehensive Evaluation: Confusion matrix, classification reports, and prediction visualizations
- Modern Web Interface: Beautiful Flask web app for real-time image classification
- CIFAR-10 Dataset: Automatically downloads and preprocesses the dataset
📊 Model Architecture
The model treats each 32x32 RGB image as a sequence of 32 rows, where each row has 96 features (32 pixels * 3 channels).
Input (Batch, 3, 32, 32)
↓
Reshape (Batch, 32, 96)
↓
Bidirectional LSTM (Hidden: 256, Layers: 2, Dropout: 0.2)
↓
Last Time Step Output
↓
Fully Connected (512) → ReLU → Dropout(0.3)
↓
Output (10 classes)
🚀 Quick Start
1. Install Dependencies
pip install -r requirements.txt
2. Train the Model
python train.py
This will:
- Download the CIFAR-10 dataset automatically
- Train the model for 50 epochs
- Save checkpoints in
./checkpoints/ - Generate training plots in
./plots/
3. Evaluate the Model
python evaluate.py
This will:
- Load the best model checkpoint
- Evaluate on the test set
- Generate confusion matrix
- Create prediction visualizations
4. Run the Web Application
python app.py
Then open your browser and navigate to http://localhost:5000
📁 Project Structure
CNN/
├── config.py # Configuration and hyperparameters
├── data_loader.py # Data loading and preprocessing
├── model.py # CNN model architecture
├── train.py # Training script
├── evaluate.py # Evaluation script
├── utils.py # Utility functions
├── app.py # Flask web application
├── requirements.txt # Python dependencies
├── templates/
│ └── index.html # Web interface HTML
├── static/
│ ├── style.css # Web interface CSS
│ └── script.js # Web interface JavaScript
├── checkpoints/ # Model checkpoints (created during training)
├── plots/ # Training visualizations (created during training)
└── data/ # CIFAR-10 dataset (downloaded automatically)
🎯 CIFAR-10 Classes
The model classifies images into 10 categories:
- Airplane
- Automobile
- Bird
- Cat
- Deer
- Dog
- Frog
- Horse
- Ship
- Truck
⚙️ Configuration
Edit config.py to customize:
- Training: epochs, batch size, learning rate
- Model: number of classes, architecture parameters
- Data: augmentation settings, normalization values
- Paths: checkpoint and plot directories
📈 Training Details
- Optimizer: SGD with momentum (0.9) and weight decay (5e-4)
- Loss Function: Cross-Entropy Loss
- Learning Rate: 0.001 with step decay
- Batch Size: 128
- Data Augmentation: Random crop and horizontal flip
- Regularization: Batch normalization and dropout
🎨 Web Interface Features
- Drag & Drop: Upload images via drag-and-drop
- Random Samples: Test with random CIFAR-10 images
- Real-time Classification: Instant predictions with confidence scores
- Top-5 Predictions: View probability distribution
- Modern UI: Dark theme with smooth animations
📊 Expected Performance
With the default configuration, the model typically achieves:
- Training Accuracy: ~90%
- Validation Accuracy: ~85%
🛠️ Requirements
- Python 3.7+
- PyTorch 2.0+
- torchvision
- Flask
- NumPy
- Matplotlib
- scikit-learn
- Pillow
- tqdm
📝 License
This project is open source and available for educational purposes.
🤝 Contributing
Feel free to fork this project and submit pull requests for improvements!
📧 Contact
For questions or feedback, please open an issue on the repository.
Built with ❤️ using PyTorch and Flask