Skip to content

Latest commit

 

History

History
 
 

oriented_rcnn

README.md

Oriented R-CNN for Object Detection

Xingxiing Xie, Gong Cheng, Jiabao Wang, Xiwen Yao, Junwei Han,

Introduction

illustration

Current state-of-the-art two-stage detectors generate oriented proposals through time-consuming schemes. This diminishes the detectors’ speed, thereby becoming the computational bottleneck in advanced oriented object detection systems. This work proposes an effective and simple oriented object detection framework, termed Oriented R-CNN, which is a general two-stage oriented detector with promising accuracy and efficiency. To be specific, in the first stage, we propose an oriented Region Proposal Network (oriented RPN) that directly generates high-quality oriented proposals in a nearly cost-free manner. The second stage is oriented R-CNN head for refining oriented Regions of Interest (oriented RoIs) and recognizing them.

Results and models

note: The ms means multiple scale image split and the rr means random rotation.

DOTA dataset

Backbone Lr schd ms rr box AP Baidu Yun Google Drive
R50-FPN 1x - - 75.87 key: v4s0 model
R101-FPN 1x - - 76.28 key: zge9 model
R50-FPN 1x √ √ 80.87 key: 66jf model
R101-FPN 1x √ √ 80.52 key: o1r6 model

HRSC2016 dataset

Backbone Lr schd ms rr voc07 voc12 Baidu Yun Google Drive
R50-FPN 3x - - 90.4 96.5 key: 02zc model
R101-FPN 3x - - 90.5 97.5 key: q3e6 model

Citation

@InProceedings{Xie_2021_ICCV,
  author = {Xie, Xingxing and Cheng, Gong and Wang, Jiabao and Yao, Xiwen and Han, Junwei},
  title = {Oriented R-CNN for Object Detection},
  booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
  month = {October},
  year = {2021},
  pages = {3520-3529} }