Fashion-MNIST: PyTorch & TensorFlow Implementations

An exploration and implementation of image classification on the Fashion-MNIST dataset using both PyTorch and TensorFlow / Keras.

This repository demonstrates model construction, training, and performance evaluation across PyTorch and TensorFlow frameworks.

---

📌 Project Overview

The project itself was originally part of a project done for ECS 170 at UC Davis. I ended up coding it in both:

• PyTorch
• TensorFlow

Consequently, the goal of this project evolved into a personal project to compare model implementation workflows and performance on the Fashion-MNIST benchmark dataset using the two distinct frameworks.

About the Dataset

Fashion-MNIST consists of 70,000 grayscale images (28x28 pixels) spanning 10 distinct clothing categories (60,000 training images and 10,000 test images).

Categories: T-shirt/top, Trouser, Pullover, Dress, Coat, Sandal, Shirt, Sneaker, Bag, Ankle boot

---

📌 Maintenance Note

Note: This codebase was originally written ~3 years ago. While the core deep learning architecture and logic remain sound, some dependencies or API calls (e.g., legacy torch export interfaces or earlier tf.keras syntax) may trigger deprecation warnings in newer framework releases.

There may also be some bugs in the current TensorFlow version. Planned Updates / Roadmap:

• Refactor PyTorch training loop to use current torch.compile and modern torch.export standards.
• Update TensorFlow dependencies and test compatibility against latest Keras versions.
• Check for and fix any possible bugs in the TensorFlow Implementation
• Add requirements.txt / environment.yml with pinned legacy versions for reproducible environments.

🛠️ Project Structure

.
├── pytorch/         # PyTorch dataset loading, model definitions, & training scripts
├── tensorflow/      # TensorFlow/Keras scripts & evaluation
└── README.md