Skip to content

LukaMilivojevic/diffusion-implementation

 
 

Repository files navigation

Diffusion-from-Scratch

Overview

This repository contains a PyTorch implementation of diffusion models, developed from first principles by two contributors. Key features include a cosine noise schedule, class-guided diffusion (without an external classifier), and training examples on CIFAR-10 and MNIST.

Features

  • Diffusion Model: Implemented from scratch in PyTorch.
  • Cosine Schedule: Noise schedule following a cosine pattern for improved image quality.
  • Class-Guided Diffusion: Directly conditioned on class labels, without requiring an external classifier.
  • Training Datasets: Trained on CIFAR-10 and MNIST.

A sample GIF demonstrating the model generating a new image is provided below.

gif of new image generation

About

No description, website, or topics provided.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Jupyter Notebook 96.7%
  • Python 3.3%