Image denoising and generation using autoencoders and diffusion models

This repository explores the use of autoencoders and diffusion models for image denoising and generation. Leveraging architectures like convolutional autoencoders and U-Net, this project compares different approaches to reduce noise in images while preserving essential features.

Features

• Denoising Autoencoders: Implementations of convolutional and U-Net-based autoencoders.
• Diffusion Models: Gradual noise-to-image transformation using diffusion models.
• Dataset: Trained on noisy images of cats (64x64, RGB) with controlled noise levels.
• Loss Functions: MSE loss functions.

Example Results

Autoencoder Results

Results of the U-Net autoencoder trained on noisy cat images for 20 epochs, with a noise level of 0.7 :

Diffusion Model Results

Epoch 50	Epoch 150

Usage

See the 'main.ipynb' notebook for a detailed overview of the project, including training and evaluation of the models.

Dependencies

See the 'requirements.txt' file for a list of dependencies.

Install dependencies:

pip install -r requirements.txt