thomascong121/aduni

Source code of "Adaptive Unified Contrastive Learning for Imbalanced Classification" presented in International Workshop on Machine Learning in Medical Imaging, 2022

AdUni

Training

For training the model, you may modify the run.sh file. Define --data_folder as the path to your dataset. We also provide our splits in the third section. To run, use bash run.sh

Testing

For testing the model, you may modify the run.sh file. Set --resume to True and define --ckpt as the path to your save model. To run, use bash run.sh

Trained Packages

We provide trained models with AdUni and different sampling technics with one train-test splits.

method	Dataset	Test acc.	Test f1	url
AdUni	ISIC2018	87.1%	79.0%	model
AdUni+upsample	ISIC2018	88.2%	79.8%	model
AdUni	APTOS2019	83.3%	69.4%	model
AdUni+upsample	APTOS2019	83.9%	71.1%	model

Reference

@inproceedings{cong2022aduni, title={Adaptive Unified Contrastive Learning for Imbalanced Classification}, author={Cong Cong, Yixing Yang, Sidong Liu, Maurice Pagnucco, Antonio DiIeva, Shlomo Berkovsky, Yang Song}, booktitle={International Workshop on Machine Learning in Medical Imaging}, year={2022}, organization={Springer} }