Publications

VB-Mitigator brings together methods introduced in the following papers. If the tool is useful in your work, please cite the framework and the specific method(s) you used (see How to cite).

VB-Mitigator (the framework)

The framework itself — the benchmarks, the mitigation-method implementations, the standardized outputs, and the app documented here.

Paper: arXiv:2507.18348 (2025) — VB-Mitigator: An open-source framework for evaluating and advancing visual bias mitigation.

@article{sarridis2025vbmitigator,
  title={VB-Mitigator: An open-source framework for evaluating and advancing visual bias mitigation},
  author={Sarridis, Ioannis and Koutlis, Christos and Papadopoulos, Symeon and Diou, Christos},
  journal={arXiv preprint arXiv:2507.18348},
  year={2025}
}

FLAC — Fairness-Aware representation learning by suppressing attribute-class associations

Learns representations that suppress the statistical association between the target label and the sensitive attribute, so the model stops relying on the protected attribute as a shortcut.

In the tool: MITIGATOR.TYPE: flac (and the blind variant flacb).
Venue: IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), 47(2):1148–1160, 2024.
Paper: IEEE Xplore.

@article{sarridis2024flac,
  title={FLAC: Fairness-aware representation learning by suppressing attribute-class associations},
  author={Sarridis, Ioannis and Koutlis, Christos and Papadopoulos, Symeon and Diou, Christos},
  journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
  volume={47},
  number={2},
  pages={1148--1160},
  year={2024},
  publisher={IEEE}
}

BAdd — Bias Mitigation through Bias Addition

Adds bias features into the model during training so the classifier is forced to become invariant to them, rather than exploiting them.

In the tool: MITIGATOR.TYPE: badd.
Venue: IEEE/CVF International Conference on Computer Vision Workshops (ICCVW), 2025, pp. 7723–7732.
Paper: IEEE Xplore.

@inproceedings{sarridis2025badd,
  title={BAdd: Bias Mitigation through Bias Addition},
  author={Sarridis, Ioannis and Koutlis, Christos and Papadopoulos, Symeon and Diou, Christos},
  booktitle={2025 IEEE/CVF International Conference on Computer Vision Workshops (ICCVW)},
  pages={7723--7732},
  year={2025},
  organization={IEEE}
}

MAVias — Mitigate Any Visual Bias

Discovers biases automatically from open-vocabulary image tags (with an LLM to judge relevance), then mitigates them — so you don’t need to know or annotate the sensitive attribute in advance.

In the tool: MITIGATOR.TYPE: mavias (and maviasb). Needs the optional [mavias] extra (pip install -e ".[mavias]").
Venue: IEEE/CVF International Conference on Computer Vision (ICCV), 2025, pp. 1271–1281.
Paper: IEEE Xplore.

@inproceedings{sarridis2025mavias,
  title={MAVias: Mitigate Any Visual Bias},
  author={Sarridis, Ioannis and Koutlis, Christos and Papadopoulos, Symeon and Diou, Christos},
  booktitle={2025 IEEE/CVF International Conference on Computer Vision (ICCV)},
  pages={1271--1281},
  year={2025},
  organization={IEEE}
}

How to cite

If you use VB-Mitigator, please cite the framework plus the method(s) you ran. For example, a study using BAdd and MAVias would cite VB-Mitigator, BAdd and MAVias. All BibTeX entries above are ready to copy.

Maintainer: Ioannis Sarridis (gsarridis@iti.gr). Work supported by the EU Horizon Europe projects MAMMOth (GA 101070285) and ELIAS (GA 101120237).