首先放上DOTA数据集官网(http://captain.whu.edu.cn/DOTAweb/index.html),官网提供水平和旋转目标提交接口,可以看到检测结果实时排行榜(http://captain.whu.edu.cn/DOTAweb/results.html),目前前五名分别来自武汉大学夏桂松团队、南京理工大学pca_lab、Cyber 公司、中科院电子所以及阿里idst。点开前面加号可以看到有的团队的方法介绍。
DOTA旋转目标赛道实时排名(2019年12-22)
以下方法介绍按照论文提交时间顺序。
时间:3 Mar 2017
题目:Arbitrary-Oriented Scene Text Detection via Rotation Proposals
链接:https://arxiv.org/abs/1703.01086
创新:
应该是第一个基于RPN架构引入旋转候选框实现任意方向的场景文本检测。基于旋转的anchor得到旋转ROI,然后提取相应特征,效果可以
pipeline
预定义anchor
时间:11 Apr 2017
题目:EAST: An Efficient and Accurate Scene Text Detector
链接:https://arxiv.org/pdf/1704.03155.pdf
知乎解读:https://zhuanlan.zhihu.com/p/37504120
创新:
时间:29 Jun 2017
题目:R2CNN: Rotational Region CNN for Orientation Robust Scene Text Detection
链接:https://arxiv.org/ftp/arxiv/papers/1706/1706.09579.pdf
知乎解读:https://zhuanlan.zhihu.com/p/41662351
创新:
时间:Sept. 2017
题目:ROTATED REGION BASED CNN FOR SHIP DETECTION
链接:https://ieeexplore.ieee.org/document/8296411
创新:
roi pooling
多任务nms
时间:26 Nov 2017
题目:Learning a Rotation Invariant Detector with Rotatable Bounding Box
链接:https://arxiv.org/pdf/1711.09405.pdf
创新:
时间:9 Jan 2018
题目:TextBoxes++: A Single-Shot Oriented Scene Text Detector
链接:https://arxiv.org/pdf/1801.02765.pdf
知乎解读:https://zhuanlan.zhihu.com/p/33723456
创新:
时间 1 Dec 2018
题目:Learning roi transformer for oriented object detection in aerial images
论文链接:https://arxiv.org/abs/1812.00155
创新:
时间:August 2018
题目:Toward arbitrary-oriented ship detection with rotated region proposal and discrimination networks
链接:https://www.researchgate.net/publication/327096241_Toward_Arbitrary-Oriented_Ship_Detection_With_Rotated_Region_Proposal_and_Discrimination_Networks
创新:
时间:17 Nov 2018
题目:SCRDet: Towards More Robust Detection for Small, Cluttered and Rotated Objects
链接:https://arxiv.org/abs/1811.07126
添加特征融合和空间、通道注意力机制。基于水平anchor,通过RPN预测粗糙ROI, 然后检测头实现对目标的任意角的坐标预测(x,y,w,h,θ),pipeline如下:
pipline
创新:
SF-Net
MDA-Net
时间:3 Mar 2019
题目:CAD-Net: A Context-Aware Detection Network for Objects in Remote Sensing Imagery
链接:https://arxiv.org/pdf/1903.00857.pdf
创新:
网络pipeline
PLCNet结构
空间注意力
时间 Aug 2019
题目:R3Det: Refined Single-Stage Detector with Feature Refinement for Rotating Object
论文链接:https://arxiv.org/abs/1908.05612
code:https://github.com/SJTU-Thinklab-Det/R3Det_Tensorflow
解读链接:https://ming71.github.io/R3Det
创新:
网络backbone使用retinanet 结构
feature refinement 模块