Fashion MNIST Classification

This project implements Convolutional Neural Networks (CNNs) to classify Fashion MNIST dataset images using two approaches: augmented and non-augmented data training.

Project Structure

• Fashion MNIST without Data Augmentation.ipynb: Basic CNN implementation
• Fashion MNIST with Data Augmentation.ipynb: Enhanced CNN with data augmentation
• Sample T-shirt image for testing the classification function

Dataset

Fashion MNIST contains 70,000 grayscale images (60,000 training, 10,000 testing) across 10 fashion categories:

labels = {
    0: "T-Shirt",
    1: "Trouser",
    2: "Pullover",
    3: "Dress",
    4: "Coat",
    5: "Sandal",
    6: "Shirt",
    7: "Sneaker",
    8: "Bag",
    9: "Ankle Boot"
}

Implementation Details

Model Architecture

• Three convolutional blocks with batch normalization
• ReLU activation and max pooling layers
• Dropout for regularization
• Fully connected layers for classification

Training Approaches

Augmented Version

Implements data augmentation with:

• Random cropping
• Horizontal flipping
• Rotation
• Color jittering

Performance:

• Validation Accuracy: 88.55%
• Training Accuracy: 85.53%
• Test Accuracy: 90.20%

Non-Augmented Version

Uses basic transformations:

• Resizing
• Normalization

Performance:

• Validation Accuracy: 93.50%
• Training Accuracy: 95.29%
• Test Accuracy: ~93%

Setup and Usage

Requirements

• Python 3.10+
• PyTorch ecosystem
• Data analysis libraries (numpy, pandas)
• Visualization tools (matplotlib, seaborn)
• Image processing (PIL)
• Machine learning utilities (scikit-learn, tqdm)

Installation

pip install torch torchvision numpy pandas matplotlib seaborn scikit-learn pillow tqdm

Image Classification

model_path = 'path/to/best_fashion_cnn.pth'
image_path = 'path/to/your/image.png'

classification = classify_image(model_path, image_path)
print(f"The image is classified as: {classification}")

Performance Analysis

Model Behavior

• Both models struggle with similar classes (Shirts vs T-shirts, Coats vs Pullovers)
• Excellent performance on distinct items (Trousers, Bags, Footwear)
• Augmented version shows better generalization despite lower raw accuracy
• Non-augmented version achieves higher accuracy but may be less robust

Recommendation

The augmented version is recommended for production deployment due to:

• Better generalization characteristics
• Improved handling of edge cases
• More robust real-world performance
• Reduced overfitting

Evaluation Metrics

Both versions include:

• Test accuracy measurements
• Detailed classification reports
• Confusion matrices
• Class-wise performance analysis

Final Verdict

While the non-augmented version showed higher raw accuracy numbers, the augmented version demonstrated better generalization characteristics and real-world applicability. The slightly lower accuracy in the augmented version is a reasonable trade-off for improved model robustness and reduced overfitting. For production deployment, the augmented version would be recommended despite its slightly lower accuracy metrics.