Skip to content

Official implementation of "Importance-Based Token Merging for Efficient Image and Video Generation" (ICCV 2025)

License

Notifications You must be signed in to change notification settings

cvlab-stonybrook/importance-token-merge

Repository files navigation

Importance-Based Token Merging for Efficient Image and Video Generation

This is the official implementation of Importance-Based Token Merging for Efficient Image and Video Generation.

International Conference on Computer Vision (ICCV), 2025

Installation

conda create -n imp_tome python=3.10
conda activate imp_tome
pip install -r requirements.txt

Demo

  • For Stable Diffusion

    python demo_sd.py
  • For PixArt-alpha

    python demo_pixart.py

Acknowledgements

  • This work was supported in part by the NASA Biodiversity Program (Award 80NSSC21K1027), and NSF Grant IIS-2212046.

  • We borrowed code from tomesd, diffusers, PixArt-alpha, zero123plus, and AnimateDiff. We thank all the authors for their great work and repos.

Citation

If you find our code useful for your research, please cite

@article{wu2024importance,
  title={Importance-Based Token Merging for Efficient Image and Video Generation},
  author={Wu, Haoyu and Xu, Jingyi and Le, Hieu and Samaras, Dimitris},
  journal={arXiv preprint arXiv:2411.16720},
  year={2024}
}

About

Official implementation of "Importance-Based Token Merging for Efficient Image and Video Generation" (ICCV 2025)

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages