This project implements a Convolutional Neural Network (CNN) using TensorFlow/Keras to classify clothing items from the Fashion MNIST dataset. The model achieves ~91% test accuracy with capabilities for:
- Image classification
- Training visualization
- Performance metrics generation
- Model checkpointing
Fashion MNIST is a benchmark dataset containing 70,000 grayscale images (28x28 pixels) across 10 fashion categories:
| Label | Class | Description |
|---|---|---|
| 0 | T-shirt/top | 👕 T-shirts, tops |
| 1 | Trouser | 👖 Pants, trousers |
| 2 | Pullover | 🧥 Sweaters |
| 3 | Dress | 👗 Dresses |
| 4 | Coat | 🧥 Jackets, coats |
| 5 | Sandal | 👡 Sandals |
| 6 | Shirt | 👔 Button-up shirts |
| 7 | Sneaker | 👟 Athletic shoes |
| 8 | Bag | 👜 Handbags |
| 9 | Ankle boot | 👢 Boots |
Dataset Split:
- Training: 60,000 images
- Testing: 10,000 images
- Grayscale (1 channel)
- Low-resolution (28×28 pixels)
- Balanced classes (6,000 images per class in training set)
- Pre-processed (centered, normalized)