This repository contains a neural network designed to classify images from the Fashion MNIST dataset. The Fashion MNIST dataset includes 60,000 training images and 10,000 test images, each a 28x28 grayscale image associated with a label from 10 classes.
The fashion_mnist_model_v2 is built using TensorFlow and Keras and contains the following layers:
- Input Layer: Flattens the 28x28 image into a 1D array.
- First Dense Layer: 512 neurons with ReLU activation and batch normalization.
- Dropout: 0.2 dropout rate for regularization.
- Second Dense Layer: 128 neurons with ReLU activation and batch normalization.
- Dropout: Another 0.2 dropout rate for regularization.
- Output Layer: 10 neurons (one for each class) with softmax activation to output probabilities.
The model uses the Adam optimizer and sparse categorical crossentropy as the loss function.
The model is trained for 20 epochs with a batch size of 32 and a validation split of 0.2 to monitor performance and prevent overfitting.
The fashion_mnist_model_v2 achieved a test accuracy of approximately 87.64%, which is a competitive score for this dataset.
To replicate the training process or to use the model for inference, you can follow these steps:
- Clone the repository:
git clone https://github.com/your-github-username/fashion-mnist-classification.git
- Navigate to the repository's directory:
cd fashion-mnist-classification
- Install the required packages:
pip install -r requirements.txt
- Run the Jupyter notebook:
jupyter notebook fashion_mnist_classification.ipynb
The training process can be visualized in the notebook, with plots for accuracy and loss over the training epochs for both training and validation sets.
- Explore more complex network architectures.
- Implement data augmentation to improve model generalization.
- Tune hyperparameters such as learning rate and batch size for better performance.
- Experiment with different optimizers and loss functions.
This project is licensed under the MIT License - see the LICENSE.md file for details.