This repository contains the implementation of simple Denoising Diffusion Probabilistic Model. DDPM uses a principle of applying a noise to an image, and then learns how to reverse this action.
Follow these steps to use this implementation:
Ensure you have Python installed on your system. This code is compatible with Python 3.9 and newer versions.
For training and testing the DDPM model, you'll need a dataset. I used CelebA dataset which you download using the following link:
After downloading, place the dataset in an appropriate directory within the your project structure, such as "./data".
-
Clone the repository to your local computer:
git clone https://github.com/dykyivladk1/DDPM.git -
Install the required dependencies. It's recommended to create and use a virtual environment:
python -m venv venv source venv/bin/activate # On Windows use `venv\Scripts\activate` pip install -r requirements.txt -
Training model
To train a model for custom dataset, you can use the following command:
python scripts/train.py --image_dir <image_dir> --device <device>