Axis-level Reflectional Symmetry Detection with Group-Equivariant Representation
Keywords: Symmetry detection, Equivariant learning, Group equivariance
Abstract: Reflectional symmetry detection remains a challenging task in machine perception, particularly in complex real-world scenarios involving noise, occlusions, and distortions. We introduce a novel equivariant approach to axis-level reflectional symmetry detection that effectively leverages dihedral group-equivariant representation to detect symmetry axes as line segments. We propose orientational anchor expansion for fine-grained rotation-equivariant analysis of diverse symmetry patterns across multiple orientations. Additionally, we develop reflectional matching with multi-scale kernels to extract effective cues of reflectional correlations, allowing for robust symmetry detection across different receptive fields. Our approach unifies axis-level detection with reflectional matching while preserving dihedral group equivariance throughout the process. Extensive experiments demonstrate the efficacy of our method while providing more accurate axis-level predictions than existing pixel-level methods in challenging scenarios.
Primary Area: applications to computer vision, audio, language, and other modalities
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.
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: 2867
Loading