Generative Dataset Distillation Based on Diffusion Model

ECCV 2024 The First Dataset Distillation Challenge Track 2 Third Place

Environment

• Python >=3.9
• Pytorch >= 1.12.1
• Torchvision >= 0.13.1
• Diffusers == 0.29.2

Generate Images

• For CIFAR100, run:

python submit_cifar100.py

• For TinyImagenet, run:

python submit_tinyimagenet.py

Note

• IPC needs to be set as a multiple of 5.

Evaluation

• Follow the official repository

Citation

If you find this paper useful for your research, please use the following BibTeX entry.

@inproceedings{su2024diffusion,
  title={Generative Dataset Distillation Based on Diffusion Model},
  author={Su, Duo and Hou, Junjie and Li, Guang and Togo, Ren and Song, Rui and Ogawa, Takahiro and Haseyama, Miki},
  booktitle={Proceedings of the European Conference on Computer Vision (ECCV), Workshop},
  year={2024}
}