CMIDD

This is the official implementation of paper Going Beyond Feature Similarity: Effective Dataset distillation based on Class-aware Conditional Mutual Information (ICLR2025) .

The repository is based on DC-DSA-DM, MTT, IDM, IDC and IID. Please cite their papers if you use the code.

Prepare pretrain models

cd pretrain
python pre_model.py

Basic experiments

cd DC-DSA-DM
python main_DM.py  --dataset CIFAR10  --ipc 10 

python main.py  --dataset CIFAR10 --ipc 10  --method DSA  --dsa_strategy color_crop_cutout_flip_scale_rotate

cd IDC
python condense.py --reproduce  -d cifar10 -f 2 --ipc 10

If you use the repo, please consider citing:

@article{zhong2024going,
  title={Going Beyond Feature Similarity: Effective Dataset distillation based on Class-aware Conditional Mutual Information},
  author={Zhong, Xinhao and Chen, Bin and Fang, Hao and Gu, Xulin and Xia, Shu-Tao and Yang, En-Hui},
  journal={arXiv preprint arXiv:2412.09945},
  year={2024}
}