PyTorch Fashion Classifier 🧥👟

An end-to-end Machine Learning pipeline built from scratch using PyTorch. This project trains and evaluates neural network architectures to classify clothing items from the Fashion MNIST dataset, showing the progression from a basic Artificial Neural Network (ANN) to a more advanced Convolutional Neural Network (CNN).

Final Test Accuracies:

• CNN Model: 93.08% 🏆
• ANN Model: 90.52%

🧠 Project Architecture

This is a modular machine learning ecosystem separated into functional components.

• Fashion_MNIST_model.py: The original basic PyTorch ANN architecture.
• Fashion_MNIST_modelCNN.py: The upgraded PyTorch CNN architecture with batch normalization.
• dataPreprocessing.py: Handles raw CSV ingestion, Train/Validation/Test splitting, and pixel normalization (Z-scaling).
• dataset.py: Manages the PyTorch DataLoader batching and hardware routing (CPU/CUDA).
• Model_Training.py: The main engine. Features a custom training loop with validation-based checkpointing. It trains the model on around 48k images, validates on 12k images, and saves the weights (.pth file) only when a new validation high score is achieved.
• plotting.py: A custom Matplotlib/Seaborn utility library for rendering image grids.
• evaluate.py: The bulk-testing script for the ANN model.
• evaluateCNN.py: The bulk-testing script for the new CNN model. Loads the saved CNN weights and evaluates them against the entire 10,000-image test dataset.
• custom_photo_verificationCNN.py: PIL Image preprocessing for custom photo classification.
• evaluateCNN.py: classifying a custom image from gallary

⚙️ Installation & Setup

1. Clone the repository:

   git clone [https://github.com/your-username/Fashion-Classifier.git](https://github.com/your-username/Fashion-Classifier.git)
   cd Fashion-Classifier

2. Install the required dependencies:

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

3. Download the Data (Required for Training): To keep this repository lightweight, the raw CSV datasets are not included.

• Go to the official Zalando Fashion MNIST Kaggle Dataset.
• Download the archive and extract fashion-mnist_train.csv and fashion-mnist_test.csv.
• Place both CSV files in the root directory of this project.

🚀 How to Run

1. Train the Model

Once your data is placed in the root folder, initiate the training loop. You can modify the import inside this file to train either the ANN:

python Model_Training.py

The script will automatically route to a CUDA GPU if available and generate a .pth file containing the trained weights.

for the CNN:

python Fashion_MNIST_modelCNN.py

2. Run Saved Models on Test Dataset

To verify the overall accuracy against the entire 10,000-image test dataset, run the evaluation script that matches your model.

For the CNN:

python evaluateCNN.py

For the ANN:

python evaluate.py

These scripts will load the saved .pth files, process the test batches without updating gradients, and output the final percentage score.

3. Classify custom images

To verify the model's strength, use custom images to see the models accuracy with real world data.

python evaluateCustomPhoto.py

🛠️ Built With

• PyTorch - Deep learning framework
• Pandas & Scikit-Learn - Data preprocessing and scaling
• Matplotlib & Seaborn - Diagnostic visualizations