Linear Gaussian Bounding Box Representation and Ring-Shaped Rotated Convolution for Oriented Object Detection

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Zhou, Zhen, Ma, Yunkai, Fan, Junfeng, Liu, Zhaoyang, Jing, Fengshui, Tan, Min
Format: Preprint
Published: 2023
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910482532663296
author Zhou, Zhen
Ma, Yunkai
Fan, Junfeng
Liu, Zhaoyang
Jing, Fengshui
Tan, Min
author_facet Zhou, Zhen
Ma, Yunkai
Fan, Junfeng
Liu, Zhaoyang
Jing, Fengshui
Tan, Min
contents In oriented object detection, current representations of oriented bounding boxes (OBBs) often suffer from boundary discontinuity problem. Methods of designing continuous regression losses do not essentially solve this problem. Although Gaussian bounding box (GBB) representation avoids this problem, directly regressing GBB is susceptible to numerical instability. We propose linear GBB (LGBB), a novel OBB representation. By linearly transforming the elements of GBB, LGBB avoids the boundary discontinuity problem and has high numerical stability. In addition, existing convolution-based rotation-sensitive feature extraction methods only have local receptive fields, resulting in slow feature aggregation. We propose ring-shaped rotated convolution (RRC), which adaptively rotates feature maps to arbitrary orientations to extract rotation-sensitive features under a ring-shaped receptive field, rapidly aggregating features and contextual information. Experimental results demonstrate that LGBB and RRC achieve state-of-the-art performance. Furthermore, integrating LGBB and RRC into various models effectively improves detection accuracy.
format Preprint
id arxiv_https___arxiv_org_abs_2311_05410
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Linear Gaussian Bounding Box Representation and Ring-Shaped Rotated Convolution for Oriented Object Detection
Zhou, Zhen
Ma, Yunkai
Fan, Junfeng
Liu, Zhaoyang
Jing, Fengshui
Tan, Min
Computer Vision and Pattern Recognition
In oriented object detection, current representations of oriented bounding boxes (OBBs) often suffer from boundary discontinuity problem. Methods of designing continuous regression losses do not essentially solve this problem. Although Gaussian bounding box (GBB) representation avoids this problem, directly regressing GBB is susceptible to numerical instability. We propose linear GBB (LGBB), a novel OBB representation. By linearly transforming the elements of GBB, LGBB avoids the boundary discontinuity problem and has high numerical stability. In addition, existing convolution-based rotation-sensitive feature extraction methods only have local receptive fields, resulting in slow feature aggregation. We propose ring-shaped rotated convolution (RRC), which adaptively rotates feature maps to arbitrary orientations to extract rotation-sensitive features under a ring-shaped receptive field, rapidly aggregating features and contextual information. Experimental results demonstrate that LGBB and RRC achieve state-of-the-art performance. Furthermore, integrating LGBB and RRC into various models effectively improves detection accuracy.
title Linear Gaussian Bounding Box Representation and Ring-Shaped Rotated Convolution for Oriented Object Detection
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2311.05410