Reliable Evaluation of Attribution Maps in CNNs: A Perturbation-Based Approach
International Journal of Computer Vision · 2025-05 · 133 (5) · 2392-2409
DOI: 10.1007/s11263-024-02282-6 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Computer Vision, online erschienen am 23.11.2024. Verfasst von Lars Nieradzik und 2 weiteren Autor:innen. Die Arbeit wurde bisher 10-mal zitiert (Crossref) und nennt 62 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Fraunhofer-Institut für Techno- und Wirtschaftsmathematik ITWM.
Abstract (Originalsprache)
In this paper, we present an approach for evaluating attribution maps, which play a central role in interpreting the predictions of convolutional neural networks (CNNs). We show that the widely used insertion/deletion metrics are susceptible to distribution shifts that affect the reliability of the ranking. Our method proposes to replace pixel modifications with adversarial perturbations, which provides a more robust evaluation framework. By using smoothness and monotonicity measures, we illustrate the effectiveness of our approach in correcting distribution shifts. In addition, we conduct the most comprehensive quantitative and qualitative assessment of attribution maps to date. Introducing baseline attribution maps as sanity checks, we find that our metric is the only contender to pass all checks. Using Kendall’s $$\tau $$ τ rank correlation coefficient, we show the increased consistency of our metric across 15 dataset-architecture combinations. Of the 16 attribution maps tested, our results clearly show SmoothGrad to be the best map currently available. This research makes an important contribution to the development of attribution maps by providing a reliable and consistent evaluation framework. To ensure reproducibility, we will provide the code along with our results.
Förderung
- Fraunhofer-Institut für Techno- und Wirtschaftsmathematik ITWM
Zitieren
APA
Nieradzik, L., Stephani, H., & Keuper, J. (2025). Reliable Evaluation of Attribution Maps in CNNs: A Perturbation-Based Approach. International Journal of Computer Vision, 133(5), 2392-2409. https://doi.org/10.1007/s11263-024-02282-6
Vancouver
Nieradzik L, Stephani H, Keuper J. Reliable Evaluation of Attribution Maps in CNNs: A Perturbation-Based Approach. International Journal of Computer Vision. 2025;133(5):2392-2409. doi:10.1007/s11263-024-02282-6
BibTeX
@article{Nieradzik2025,
author = {Nieradzik, Lars and Stephani, Henrike and Keuper, Janis},
title = {Reliable Evaluation of Attribution Maps in CNNs: A Perturbation-Based Approach},
journal = {International Journal of Computer Vision},
year = {2025},
volume = {133},
number = {5},
pages = {2392--2409},
doi = {10.1007/s11263-024-02282-6}
}
EUVIMED