ljj7975/pytorch-ddpm

pytorch-ddpm

Simple Denoising Diffusion Probabilistic Models (DDPM) implementation in PyTorch.

The base implementation is taken from: https://github.com/cloneofsimo/minDiffusion.

Modifications that I have made on top includes

• class bias
• positional encoding
• residual connection to the model

In order to generate diffusion model for mnist, simply run

$ python train_mnist.py

sample MNIST outputs are available here

prediction: 6, 1, 3, 3, 1, 1, 1, 1, 6, 5, 0, 8, 4, 3, 8, 7

Please note that train_celelba.py and train_cifar10.py are not yet working