cnn

CIFAR-10 CNN Classifier

An end-to-end Convolutional Neural Network (CNN) implementation for classifying images from the CIFAR-10 dataset using PyTorch.

Project Structure

CNN/
├── main.py              # Main entry point
├── requirements.txt     # Project dependencies
├── src/
│   ├── data.py         # Data loading utilities
│   ├── model.py        # CNN model architecture
│   ├── train.py        # Training script
│   └── evaluate.py     # Evaluation script
└── model.pth           # Saved model (after training)

Setup

1. Install dependencies:

pip install -r requirements.txt

Usage

Option 1: Using the main script (Interactive)

python main.py

Option 2: Direct training

cd src
python train.py

Option 3: Direct evaluation

cd src
python evaluate.py

Model Architecture

The CNN consists of:

• 2 convolutional blocks (each with 2 conv layers)
• Max pooling layers
• Fully connected layers with dropout
• Output: 10 classes (CIFAR-10 categories)

CIFAR-10 Classes

1. Airplane
2. Automobile
3. Bird
4. Cat
5. Deer
6. Dog
7. Frog
8. Horse
9. Ship
10. Truck

Training Details

• Optimizer: Adam
• Loss Function: Cross-Entropy Loss
• Default Epochs: 10
• Batch Size: 64
• Learning Rate: 0.001

The best model (based on test accuracy) is automatically saved as model.pth.

Expected Performance

With the default configuration, you should expect:

• Training accuracy: ~70-80% after 10 epochs
• Test accuracy: ~65-75% after 10 epochs

For better performance, consider:

• Training for more epochs (20-30)
• Data augmentation
• Learning rate scheduling
• More complex architectures