DM-NonUniform

Official code for Accelerating Diffusion Sampling with Optimized Time Steps (CVPR 2024)

Accelerating Diffusion Sampling with Optimized Time Steps (CVPR 2024)
Shuchen Xue, Zhaoqiang Liu†, Fei Chen, Shifeng Zhang, Tianyang Hu, Enze Xie, Zhenguo Li
University of Chinese Academy of Sciences, University of Electronic Science and Technology of China, Huawei Noah’s Ark Lab

---

Abstract

Discretization of sampling time steps are mainly hand-crafted designed, such as uniform-t scheme, quadratic-t scheme, uniform logSNR scheme and EDM scheme. We propose an optimization-based method to choose appropriate time steps for a specific numerical ODE solver for Diffusion Models.

Integration with DPM-Solver, UniPC

Add the following code in "get_time_steps" method in DPM-Solver or UniPC

from step_optim import StepOptim

elif skip_type == "optimized":
    optimizer = StepOptim(self.noise_schedule)
    t, _ = optimizer.get_ts_lambdas(N, t_0, optimized_type)
    t = t.to(device).to(torch.float32)
    return t

For pixel-space diffusion models, we recommend use "optimized_type" as "unif", which means that the optimization algorithm will use uniform-logSNR steps as initialization; for latent-space diffusion models, we recommend use "optimized_type" = "unif_t", which means that the optimization algorithm will use uniform-time steps as initialization.

Citation

If you find our work useful in your research, please consider citing:

@article{xue2024accelerating,
  title={Accelerating Diffusion Sampling with Optimized Time Steps},
  author={Xue, Shuchen and Liu, Zhaoqiang and Chen, Fei and Zhang, Shifeng and Hu, Tianyang and Xie, Enze and Li, Zhenguo},
  journal={arXiv preprint arXiv:2402.17376},
  year={2024}
}