narjesmath / LEVIR-Ship

This is the official release of LEVIR-Ship, which is a dataset for tiny ship detection under medium-resolution remote sensing images

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

LEVIR-Ship: A Large-Scale Tiny Ship Detection Dataset under Medium-Resolution Remote Sensing Images

visitors GitHub stars CC BY 4.0

Share us a ⭐ if this repo does help

This is the official release of the LEVIR-Ship dataset in "A Degraded Reconstruction Enhancement-based Method for Tiny Ship Detection in Remote Sensing Images with A New Large-scale Dataset" [IEEE | Lab Server]. (Accepted by TGRS 2022)

If you encounter any question, please feel free to contact us. You can create an issue or just send email to me windvchen@gmail.com. Also welcome for any idea exchange and discussion.

Updates

06/11/2022

The LEVIR-Ship dataset and some useful tools are released.

06/10/2022

We will complete the arrangement of the dataset and its related tools within three days and make it public here. Please be patient.

Table of Contents

Introduction

LEVIR-Ship

Images in LEVIR-Ship are captured from multispectral cameras of GaoFen-1 and GaoFen-6 satellites with a spatial resolution of 16m. We only use the R, G, and B bands. 85 scenes have been collected in the dataset with pixel resolutions between 10000×10000 and 50000×20000. We crop the original images to finally get 1973 positive samples and 1923 negative samples with the size of 512×512.

To our most knowledge, LEVIR-Ship is the first public tiny ship detection dataset specific to medium-resolution remote sensing images.

Download Source

For the whole dataset

  • All Annotations: [Google Drive | Baidu Pan (code:6pr3)]

    The annotations are in Yolo format, i.e., [class, x_center, y_center, width, height]. If you need COCO format, you can refer to the following provided transformation tools.

  • All Images: [Google Drive | Baidu Pan (code:s1m9)]

For the partitioned dataset used in DRENet

YOLO Format

COCO Format

Attribute

Compared with other datasets

Dataset Images Number Instances Number Source Resolution Year
NWPU VHR-10 57 302 Google Earth 0.5-2m 2014
HRSC2016 1070 2976 Google Earth 0.4-2m 2016
DIOR 2702 62400 Google Earth 0.5-30m 2018
HRRSD 2165 3886 Google Earth & Baidu Earth 0.15-1.2m 2019
LEVIR-Ship 3896 3219 GaoFen-1 & GaoFen-6 16m 2021

Examples of different situations

Images in LEVIR-Ship show very different conditions, due to the influence of different time, different photographers, and different locations, which can bring many challenges to the tiny ship detection task. In other words, LEVIR-Ship is much closer to the real world situations, compared to some Google Earth-based datasets whose images are usually clean and cloudless. Thus, LEVIR-Ship can bring networks better generalization, stability, and sufficiency.

Different amount of cloud Cloud

Different percentages of land coverage Land

Different light intensity Light

Different sea area characteristics Sea surface

Tools

In this repo, we provide some useful tools for dataset visualization and process.

Citation

If you make use of LEVIR-Ship dataset in your research, please cite:

@ARTICLE{9791363,
  author={Chen, Jianqi and Chen, Keyan and Chen, Hao and Zou, Zhengxia and Shi, Zhenwei},
  journal={IEEE Transactions on Geoscience and Remote Sensing},
  title={A Degraded Reconstruction Enhancement-based Method for Tiny Ship Detection in Remote Sensing Images with A New Large-scale Dataset},
  year={2022},
  volume={},
  number={},
  pages={1-1},
  doi={10.1109/TGRS.2022.3180894}}

License

The codes of this project is licensed under the MIT License. See LICENSE for details.

The annotations in the LEVIR-Ship dataset are licensed under a Creative Commons Attribution 4.0 License.

CC BY 4.0

We do not own the copyright of the images. Use of the images must abide by China Centre For Resources Satellite Data and Application Terms of Use.

About

This is the official release of LEVIR-Ship, which is a dataset for tiny ship detection under medium-resolution remote sensing images

License:MIT License


Languages

Language:Python 100.0%