FakE ?!

FakeE is a DDPM with a U-Net backbone, trained on MNIST.

A simple diffusion model implemented in PyTorch.

Requirements

pip install -r requirements.txt

Train

python train.py

Generate

python test.py

The model starts from random noise and progressively denoises it to generate a new MNIST digit.

Project Structure

FakeE/
├── model/
│   └── FakeE.pth
├── model.py
├── train.py
├── test.py
├── requirements.txt
└── README.md