Diffusion Models with SDEs

This is an attempt to implement diffusion based generative models (Score-based models, DDPMs) using Stochastic Differential Equations from the paper Score-Based Generative Modeling through Stochastic Differential Equations

What is in here?

• Variance Preserving SDE
• Probability Flow sampling
• MNIST, CIFAR 10, 100 with a simple UNet model as score function (time dependence is not yet implemented)

Training

python main.py --c 'configs/vpsde_mnist.yaml'

Next steps..

• Implement Variance Exploding SDE, Sub-VPSDE
• Implement sampling by solving reverse SDE
• Maximum likelihood based training using likelihood weighting (from this paper)

Samples