Gotta Go Fast When Generating Data with Score-Based Models

This repo contains the official implementation for the paper Gotta Go Fast When Generating Data with Score-Based Models, which shows how to generate data as fast as possible with score-based models using a well-designed SDE solver. See the blog post for more details.

This code is a heavy modification of the Generative Modeling through Stochastic Differential Equations repository.

To run the experiments in the paper

See the requirements. Change the settings and folders in https://github.com/AlexiaJM/score_sde_fast_sampling/blob/main/experiments.sh and run parts of the script to run the CIFAR-10, LSUN-Church, and FFHQ experiments.

The SDE solver can be found here and the loop here.

For general usage

Please refer to the original code.

Pretrained checkpoints

https://drive.google.com/drive/folders/10pQygNzF7hOOLwP3q8GiNxSnFRpArUxQ?usp=sharing

References

If you find the code useful for your research, please consider citing

@article{jolicoeurmartineau2021gotta,
      title={Gotta Go Fast When Generating Data with Score-Based Models}, 
      author={Alexia Jolicoeur-Martineau and Ke Li and R{\'e}mi Pich{\'e}-Taillefer and Tal Kachman and Ioannis Mitliagkas},
      journal={arXiv preprint arXiv:2105.14080},
      year={2021}
}

and

@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}
}

Official theme song can be found here: https://soundcloud.com/emyaze/gotta-go-fast.

Samples (see the paper for more samples)