---

MNIST & Fashion MNIST Classification with PyTorch

This repository contains PyTorch implementations for training, evaluating, and deploying simple feed-forward neural networks on the MNIST and Fashion MNIST datasets. It includes:

1. MNIST – Handwritten digit classification.
2. Fashion MNIST (fmnist) – Clothing article classification (basic training and evaluation).
3. Fashion MNIST v2 (fmnist2) – Fashion MNIST classification with extended code and testing.
4. Streamlit App (appfmnist) – Interactive web app to upload images and predict Fashion MNIST classes.

---

📂 Project Structure

.
├── mnist.py        # MNIST training, evaluation, and visualization
├── fmnist.py       # Fashion MNIST training, evaluation, and visualization
├── fmnist2.py      # Fashion MNIST variant with additional test image visualization
├── appfmnist.py    # Streamlit app for Fashion MNIST predictions
├── data/           # MNIST/Fashion MNIST datasets (downloaded automatically)
├── fashion_mnist_model.pth # Saved trained model weights (generated after training)

---

📦 Requirements

Install dependencies:

pip install torch torchvision streamlit pillow matplotlib numpy

---

🖋 How to Run

1️⃣ MNIST

Train and evaluate on MNIST:

python mnist.py

• Downloads MNIST dataset automatically.
• Trains a simple feed-forward neural network.
• Prints accuracy on test dataset.
• Displays sample predictions.

---

2️⃣ Fashion MNIST (fmnist)

Train and evaluate on Fashion MNIST:

python fmnist.py

• Trains a neural network on Fashion MNIST.
• Saves trained weights to fashion_mnist_model.pth.
• Evaluates accuracy.
• Displays predictions on sample test images.

---

3️⃣ Fashion MNIST v2 (fmnist2)

Alternate Fashion MNIST pipeline with sample/random image testing inside Streamlit:

python fmnist2.py

• Loads trained weights.
• Allows testing on random Fashion MNIST images or uploaded images.

---

4️⃣ Fashion MNIST Streamlit App (appfmnist)

Run the web app:

streamlit run appfmnist.py

• Upload your own .jpg, .png, .jpeg images.
• See predicted class in real time.
• Uses the model trained in fmnist.py.

---

🧠 Model Architecture

All models use the same simple feed-forward architecture:

Input Layer:  784 nodes (28x28 pixels)
Hidden Layer1: 128 nodes (ReLU)
Hidden Layer2: 64 nodes (ReLU)
Output Layer:  10 nodes (class scores)

---

📊 Datasets

• MNIST – Handwritten digits (0–9).
• Fashion MNIST – Clothing items (T-shirt/top, Trouser, Pullover, Dress, Coat, Sandal, Shirt, Sneaker, Bag, Ankle boot).

Both datasets are loaded using:

from torchvision import datasets, transforms

and normalized to range [-1, 1].

---

🖼 Sample Predictions

Example output from MNIST/Fashion MNIST scripts:

Image	Predicted
🖼 Digit "7"	7
👕 T-shirt/top	T-shirt/top

---

🚀 Future Improvements

• Add convolutional neural network (CNN) for higher accuracy.
• Support batch image predictions in the Streamlit app.
• Deploy app on cloud services (e.g., Streamlit Cloud, Heroku).

---

📜 License

This project is licensed under the MIT License.

---