MNIST Autoencoder

This repository contains a simple convolutional autoencoder implemented in PyTorch for reconstructing images from the MNIST dataset. The model learns to compress and reconstruct handwritten digit images using convolutional and transposed convolutional layers.

Features

• Loads the MNIST dataset.
• Implements an Encoder to reduce image dimensionality.
• Implements a Decoder to reconstruct images.
• Trains using Mean Squared Error (MSE) loss and the Adam optimizer.
• Displays sample reconstructions before and after training.

Model Architecture

Encoder

The encoder consists of a series of convolutional layers that progressively downsample the input image:

• Conv2D (1 → 8 channels, kernel=15, stride=1, padding=1) → Output: 8×16×16
• Conv2D (8 → 16 channels, kernel=3, stride=2, padding=1) → Output: 16×8×8
• Conv2D (16 → 32 channels, kernel=3, stride=2, padding=1) → Output: 32×4×4
• Conv2D (32 → 64 channels, kernel=5, stride=2, padding=1) → Output: 64×1×1

Decoder

The decoder reverses the encoding process using transposed convolutions:

• ConvTranspose2D (64 → 32 channels, kernel=4, stride=1, padding=0) → Output: 32×4×4
• ConvTranspose2D (32 → 16 channels, kernel=2, stride=2, padding=0) → Output: 16×8×8
• ConvTranspose2D (16 → 8 channels, kernel=2, stride=2, padding=0) → Output: 8×16×16
• ConvTranspose2D (8 → 1 channel, kernel=4, stride=2, padding=3) → Output: 1×28×28

Training

• The model is trained for 15 epochs using a batch size of 32.
• Uses Mean Squared Error (MSE) loss function.
• Optimized using the Adam optimizer with a learning rate of 1e-2.
• Training loop iterates over batches, performing forward and backward propagation to minimize reconstruction error.

Results

Before Training

The untrained model produces blurry or random reconstructions of MNIST digits.

After Training

The trained autoencoder successfully reconstructs MNIST images with high fidelity.

How to Use

Requirements

• Python 3.x
• PyTorch
• torchvision
• matplotlib

Running the Code

1. Clone the repository:

   git clone https://github.com/your-username/mnist-autoencoder.git
   cd mnist-autoencoder

2. Install dependencies:

   pip install torch torchvision matplotlib

3. Run the script to train and visualize the results:

   python autoencoder.py

Future Improvements

• Add batch normalization and dropout for better generalization.
• Experiment with different architectures (e.g., Variational Autoencoders).
• Extend to color images and other datasets.