Visual Explanations via Iterated Integrated Attributions

Oren Barkan, ‪Yehonatan Elisha‬‏, Yuval Asher, Amit Eshel, Noam Koenigstein

ICCV 2023 pp. 2073-2084

doi:10.1109/ICCV51070.2023.00198 /iccv/2023/barkan2023iccv-visual/

Abstract

We introduce Iterated Integrated Attributions (IIA) - a generic method for explaining the predictions of vision models. IIA employs iterative integration across the input image, the internal representations generated by the model, and their gradients, yielding precise and focused explanation maps. We demonstrate the effectiveness of IIA through comprehensive evaluations across various tasks, datasets, and network architectures. Our results showcase that IIA produces accurate explanation maps, outperforming other state-of-the-art explanation techniques.

PDF ICCV Semantic Scholar

Cite

Text

Barkan et al. "Visual Explanations via Iterated Integrated Attributions." International Conference on Computer Vision, 2023. doi:10.1109/ICCV51070.2023.00198

Markdown

[Barkan et al. "Visual Explanations via Iterated Integrated Attributions." International Conference on Computer Vision, 2023.](https://mlanthology.org/iccv/2023/barkan2023iccv-visual/) doi:10.1109/ICCV51070.2023.00198

BibTeX

@inproceedings{barkan2023iccv-visual,
  title     = {{Visual Explanations via Iterated Integrated Attributions}},
  author    = {Barkan, Oren and Elisha‬‏, ‪Yehonatan and Asher, Yuval and Eshel, Amit and Koenigstein, Noam},
  booktitle = {International Conference on Computer Vision},
  year      = {2023},
  pages     = {2073-2084},
  doi       = {10.1109/ICCV51070.2023.00198},
  url       = {https://mlanthology.org/iccv/2023/barkan2023iccv-visual/}
}