Imbalaced Classification and Robust Semantic Segmentation

This repo implements two algoritms.

1. The imbalance clibration (IC) algorithm for image classification. Please go to the classification folder for IC.
2. The uncertainty-aware noisy-or multi-modal fusion and imbalance calibration (UNO-IC) algorithm for semantic segmentation, UNO-IC. Please go to the segmentation folder for UNO-IC.

Acknowledgments

• This work was supported by ONR grant N00014-18-1-2829.

Paper

1. Posterior Re-calibration for Imbalanced Datasets
2. UNO: Uncertainty-aware Noisy-Or Multimodal Fusion for Unanticipated Input Degradation

Citation

If you find this repository useful, please cite our paper(s):

@inproceedings{tian2020posterior,
    title={Posterior Re-calibration for Imbalanced Datasets},
    author={Junjiao Tian, Yen-Cheng Liu,  Nathaniel Glaser, Yen-Chang Hsu, and Zsolt Kira},
    booktitle={Proceedings of the 34th Conference on Neural Information Processing Systems},
    year={2020}
}
@inproceedings{tian2019uno,
    title={UNO: Uncertainty-aware Noisy-Or Multimodal Fusion for Unanticipated Input Degradation},
    author={Junjiao Tian, Wesley Cheung,  Nathaniel Glaser, Yen-Cheng Liu, and Zsolt Kira},
    booktitle={Proceedings of the  IEEE International Conference on Robotics and Automation (ICRA)},
    year={2020}
}