Rotation-Equivariance and Position Encodings for Enhancing Local Descriptors
Primary Area: representation learning for 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.
Keywords: Rotation equivariance, keypoint extraction
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
TL;DR: A robust local feature point extractor with strong rotational invariance.
Abstract: Keypoint extraction and description are crucial issues in robot vision. In recent years, deep learning based keypoint extraction have exhibited robustness to variations in lighting and viewpoint. However, due to the lack of rotational invariance in traditional convolutional networks, performance of deep learning-based keypoint significantly deteriorates under large rotations. Group-equivariant neural networks based Keypoint address the issue of rotational equivariance, but their overall performance also suffers. This paper addresses the problem from the perspective of keypoint description and proposes a fusion of locally rotation-equivariant descriptions with globally encoded positional information and a directional uncertainty weighted descriptor loss. This effectively enhances the performance of keypoint extraction and description. Validation is conducted on rotated-HPatches, rotated-MegaDepth and rotated-YFCC100M datasets.
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: 4533
Loading