EVA

This repo contains the Pytorch implementation of our paper:

Evolution-aware VArance (EVA) Coreset Selection for Medical Image Classification

Yuxin Hong, Xiao Zhang, Xin Zhang, Joey Tianyi Zhou.

• Accepted at ACMMM 2024 as Oral Presentation.
• 🎉 Nominated for Best Paper Awards.

💻 Installation and Requirements

We strongly recommend that you create your own virtual environment first to avoid conflicts. Then setup the required Python environments:

pip install requirements.txt

Dataset

We validate the effectiveness of EVA mainly on MedMNIST. Install medmnist as a standard Python package from PyPI:

pip install medmnist

Or install from source:

pip install --upgrade git+https://github.com/MedMNIST/MedMNIST.git

Check whether you have installed the latest code version:

>>> import medmnist
>>> print(medmnist.__version__)

Please download .npz files from here to /.medmnist directory.

🏃‍♀️ Getting Started

Train Classifiers on the Entire Dataset

This step is necessary to collect training dynamics for future coreset selection.

python train.py --dataset organsmnist --gpuid 0 --epochs 200 --lr 0.1 --network resnet18 --batch_size 256 --task_name all-data --base_dir ./data-model/organsmnist --download --as_rgb

After completing this step, you will obtain a .pickle file under the path ./data-model/organsmnist/all-data, which will be used for sample importance calculation.

Sample Importance Score Calculation

We need to first calcualte the different importance scores for coreset selection.

python generate_importance_score.py --dataset organsmnist --gpuid 0 --base_dir ./data-model/organsmnist --e_min 0 --e_max 10 --l_min 100 --l_max 110 --task_name all-data --as_rgb

After the calculation, you will obtain a .pickle file storing sorted sample indexes and their respective importance scores.

Train Classifiers on the Selected Coreset

Here we use 30% selection rate on OrganSMNIST as an example.

python train.py --dataset organsmnist --gpuid 0 --epochs 200 --base_dir ./data-model/organsmnist/eva --coreset --coreset_mode coreset --data-score-path ./data-model/organsmnist/all-data/data-score-all-data-0-10-100-110.pickle --coreset_key eva --as_rgb --task_name eva-0-10-100-110-0.3  --coreset_ratio 0.3

• For an aggressive low selection rate, pls set a smaller batch size for better performance.

👍 Citation

If you find our work useful for your research, please cite our paper:

@inproceedings{Hong2024EVA,
    author = {Hong, Yuxin and Zhang, Xiao and Zhang, Xin and Zhou, Joey Tianyi},
    title = {Evolution-aware VAriance (EVA) Coreset Selection for Medical Image Classification},
    year = {2024},
    booktitle = {Proceedings of the 32nd ACM International Conference on Multimedia},
    pages = {301–310},
    location = {Melbourne VIC, Australia},
}