ag027592 / unc-net

Code to support characterizing sources of uncertainty to proxy calibration and disambiguate annotator and data bias.

This code is built upon an a TensorFlow implementation of "FaceNet: A Unified Embedding for Face Recognition and Clustering" where a baseline classifier is augmented by uncertainty quanitifaction capabilities.

For more information, see here.

Reference

If you use uncertainty quantification aspects of this code, please reference the following paper:

@article{ghandeharioun2019characterizing,
    title={Characterizing Sources of Uncertainty to Proxy Calibration and Disambiguate Annotator and Data Bias},
    author={Asma Ghandeharioun and Brian Eoff and Brendan Jou and Rosalind W. Picard},
    year={2019},
    eprint={1909.09285},
    archivePrefix={arXiv},
    primaryClass={cs.LG},
    note = {2019 ICCV Workshop on Interpreting and Explaining Visual Artificial Intelligence Models}
}