Learning Rotation-Invariant Representation using Rotation-Equivariant CNNs

Sangjun Han, Kwanghyun Ko, Woojin Cheong, Myungjoo Kang

Keywords: Self-supervised Learning, Contrastive Learning, Equivariance, Rotation-Invariance, Guiding Invariance with Equivariance

Abstract: Conventional self-supervised learning (SSL) methods, such as SimCLR and SimSiam, have demonstrated significant effectiveness. However, their feature representation is not robust to image rotations, as rotational augmentation may negatively impact the framework. In this paper, we address this limitation by applying SSL to group-equivariant CNNs, specifically rotation-equivariant CNNs, to develop robust features. To learn expressive, rotation-invariant features, we introduce our training method, Guiding Invariance with Equivariance (GIE), which simultaneously trains both invariant features and the equivariance score for images. The equivariance score guides the rotation-equivariant features through an attention-weighted sum mechanism, enabling the development of rotation-invariant features. Through experiments, we demonstrate that our GIE method not only extracts high-performing features under four discrete rotations but also achieves robustness to random-degree rotations through rotation augmentation training. These results highlight the effectiveness of our method in achieving robust rotation-invariance.

Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation learning

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 13421