Just a Matter of Scale? Reevaluating Scale Equivariance in Convolutional Neural Networks

Altstidl TR, Nguyen A, Schwinn L, Köferl F, Mutschler C, Eskofier B, Zanca D (2023)

Publication Language: English

Publication Type: Conference contribution, Original article

Publication year: 2023

Pages Range: 1-8

Conference Proceedings Title: Proc. Intl. Joint Conf. Neural Netw. (IJCNN)

Event location: Gold Coast, Australia AU

DOI: 10.1109/IJCNN54540.2023.10191724

Abstract

The widespread success of convolutional neural networks may largely be attributed to their intrinsic property of translation equivariance. However, convolutions are not equivariant to variations in scale and fail to generalize to objects of different sizes. Despite recent advances in this field, it remains unclear how well current methods generalize to unobserved scales on real-world data and to what extent scale equivariance plays a role. To address this, we propose the novel Scaled and Translated Image Recognition (STIR) benchmark based on four different domains. Additionally, we introduce a new family of models that applies many re-scaled kernels with shared weights in parallel and then selects the most appropriate one. Our experimental results on STIR show that both the existing and proposed approaches can improve generalization across scales compared to standard convolutions. We also demonstrate that our family of models is able to generalize well towards larger scales and improve scale equivariance. Moreover, due to their unique design we can validate that kernel selection is consistent with input scale. Even so, none of the evaluated models maintain their performance for large differences in scale, demonstrating that a general understanding of how scale equivariance can improve generalization and robustness is still lacking.

Authors with CRIS profile

Thomas Robert Altstidl Machine Learning and Data Analytics Lab An Nguyen Machine Learning and Data Analytics Lab Leo Schwinn Machine Learning and Data Analytics Lab Franz Köferl Machine Learning and Data Analytics Lab Christopher Mutschler Department Artificial Intelligence in Biomedical Engineering (AIBE) Björn Eskofier Machine Learning and Data Analytics Lab Dario Zanca Machine Learning and Data Analytics Lab

How to cite

APA:

Altstidl, T.R., Nguyen, A., Schwinn, L., Köferl, F., Mutschler, C., Eskofier, B., & Zanca, D. (2023). Just a Matter of Scale? Reevaluating Scale Equivariance in Convolutional Neural Networks. In Proc. Intl. Joint Conf. Neural Netw. (IJCNN) (pp. 1-8). Gold Coast, Australia, AU.

MLA:

Altstidl, Thomas Robert, et al. "Just a Matter of Scale? Reevaluating Scale Equivariance in Convolutional Neural Networks." Proceedings of the 2023 International Joint Conference on Neural Networks (IJCNN), Gold Coast, Australia 2023. 1-8.

BibTeX: Download