DDPM SDE

A minimal PyTorch implementation of Score-Based Generative Modeling through Stochastic Differential Equations.

Here are some images generated by the model (Sub-VP SDE) trained on CIFAR-10 for about 310K steps, using the Euler–Maruyama predictor:

This implementation is heavily based on the official implementation. Currently, it features only the Sub-VP SDE with the Euler–Maruyama predictor and no corrector. As for the architectural changes to the U-Net, I've implemented only the scaling of skip connections.

---

Dependencies

• numpy
• pytorch
• einops

---

References

@inproceedings{
    song2021scorebased,
    title={Score-Based Generative Modeling through Stochastic Differential Equations},
    author={Yang Song and Jascha Sohl-Dickstein and Diederik P Kingma and Abhishek Kumar and Stefano Ermon and Ben Poole},
    booktitle={International Conference on Learning Representations},
    year={2021},
    url={https://openreview.net/forum?id=PxTIG12RRHS}
}