Skip to content

Sampling-Balance based Multi-stage Network (SB-MSN) for aerial image object detection

Notifications You must be signed in to change notification settings

weihancug/Sampling-Balance_Multi-stage_Network

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

5 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Sampling-Balance_Multi-stage_Network

This code is to replicate the expeirments from the paper Improving Training Instance Quality in Aerial Image Object Detection with A Sampling-balance based Multi-stage Network

1. Introduction

This network is built based on Pytorch 1.1 and MMdetection v1.0rc1

Please refer to INSTALL.md to install the MMdetection framework. It should be noted that a correct version of MMdetection should be download first, which can ensure the codes to be executable.

You should first prepare the used dataset as VOC format.

Three datasets NWPU-VHR10, DIOR, HRRSD are implemented in the branch. If you use one of them, you can set

dataset_type = 'VOCDataset' # for NWPU10

dataset_type = 'HRRSD_Dataset' # for HRRSD dataset_type = 'DOIR_Dataset' # for DIOR

The codes can repreduce the expeiremnts in the paper.

2. The overall architecture of the proposed detector.

method image

3. Some prediction examples of the proposed method on the NWPU VHR-10 data set (Green boxes are the correct predictions. Blue boxes are the false predictions. Red boxes are the missing predictions).

NWPU image

Citation

If you use this method in your research, please cite this paper.

@article{SBNet,
  title   = {Improving Training Instance Quality in Aerial Image Object 
             Detection With a Sampling-Balance-Based Multistage Network},
  author  = {Wei Han, Runyu Fan, Lizhe Wang, Ruyi Feng, Fengpeng Li, 
             Ze Deng, and Xiaodao Chen},
  journal = {{IEEE} Trans. Geosci. Remote. Sens.}, 
  doi     = {10.1109/TGRS.2020.3038803},
  year    = {2020},
  pages   = {1-15}
}

About

Sampling-Balance based Multi-stage Network (SB-MSN) for aerial image object detection

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published