作者：Xue Yang, Junchi Yan, Ziming Feng, Tao He
**机构：**上海交通大学
发布时间：2019
发布期刊：Thirty-Five AAAI Conference on Artificial Intelligence (AAAI2021)
论文地址：https://arxiv.org/abs/1908.05612
代码：
Rotation detection is a challenging task due to the diffificulties of locating the multi-angle objects and separating them effectively from the background. Though considerable progress has been made, for practical settings, there still exist challenges for rotating objects with large aspect ratio, dense distribution and category extremely imbalance. In this paper, we propose an end-to-end refifined single-stage rotation detector for fast and accurate object detection by using a progressive regression approach from coarse to fifine granularity. Considering the shortcoming of feature misalignment in existing refifined single stage detector, we design a feature refifinement module to improve detection performance by getting more accurate features. The key idea of feature refifinement module is to re-encode the position information of the current refifined bounding box to the corresponding feature points through pixel-wise feature interpolation to realize feature reconstruction and alignment. For more accurate rotation estimation, an approximate SkewIoU loss is proposed to solve the problem that the calculation of SkewIoU is not derivable. Experiments on three popular remote sensing public datasets DOTA, HRSC2016, UCAS-AOD as well as one scene text dataset ICDAR2015 show the effectiveness of our approach. Tensorflflow and Pytorch version codes are available at https://github.com/Thinklab-SJTU/R3Det_Tensorflow and https://github.com/SJTU-Thinklab-Det/r3det-on-mmdetection, and R3*Det is also integrated in our open source rotation detection benchmark: https://github.com/yangxue0827/RotationDetection.*
目前通用的horizontal detectors 在许多实际应用中有基本的局限性，例如：
目前，检测旋转物体目前存在的问题
目前，已有的检测算法可以分为双阶段检测算法和单阶段检测算法
而单阶段算法，为了追求其速度，则不能使用上Roi这种速度，因此，本文提出了feature refinement module来解决这个问题
本文贡献：