Spacecraft-DS: A Spacecraft Dataset for Key Components Detection and Segmentation via Hardware-in-the-loop Capture

News

• 2022.11.23 Update 125 Spacecraft images.
• 2022.9.26 Our paper is submmited to IEEE Sensors Journa!

Spacecraft-DS

• Spacecraft-DS contains 2 types of spacecrafts, 9 types of key components, 3 types of illuminations (normal illumination, low illumination, and high saturation), and 2 types of motion states (approaching phase and hovering phase)
  
  
• Spacecraft-DS can be downloaded here
• Google Drive

---

Spacecraft detection and segmentation models

• The Spacecraft detection and segmentation models in our paper could be downloaded here

• Detection

• Google Drive

• Segmentation

• Google Drive

• If you can not download, please contact [email protected]

---

For more information, please check our paper(Spacecraft-DS: A Spacecraft Dataset for Key Components Detection and Segmentation via Hardware-in-the-loop Capture)

Abstract

The detection and segmentation of key spacecraft components is a crucial prerequisite for the successful execution of on-orbit capture tasks. However, the existing spacecraft datasets are plagued by several problems. These include a lack of a unified dataset for component detection and segmentation, a lack of consideration of the motion states of the spacecraft, and a lack of consideration of extreme illuminations. These problems hinder the development of related research. In response to the above problems, this paper aims to bridge this gap by releasing a dataset for detecting and segmenting key spacecraft components. In contrast to synthetic images, the proposed dataset is built via hardware-in-the-loop capture. The dataset contains 2 types of spacecrafts, 9 types of key components, 3 types of illuminations (normal illumination, low illumination, and high saturation), and 2 types of motion states (approaching phase and hovering phase). Furthermore, we conduct extensive experiments on Spacecraft-DS with state-of-the-art methods for detection and segmentation tasks and provide comprehensive analysis. In addition, the dataset can also be used to evaluate many computer vision tasks, such as small object detection, etc.

Detection Benchmark(AP)

Model	2	3	5	11	12	13	14	15	17
Faster R-CNN	0.607	0.708	0.644	0.629	0.884	0.696	0.792	0.847	0.708
Cascade R-CNN	0.609	0.725	0.665	0.632	0.923	0.713	0.815	0.87	0.747
Grid R-CNN	0.602	0.702	0.648	0.621	0.874	0.707	0.787	0.823	0.724
Sparse R-CNN	0.561	0.702	0.618	0.574	0.920	0.668	0.798	0.896	0.700
SSD	0.565	0678	0.614	0.595	0.835	0.674	0.749	0.813	0.653
YOLOV3	0.450	0.500	0.433	0.475	0.677	0.494	0.357	0.553	0.423
YOLOF	0.541	0.671	0.572	0.536	0.847	0.633	0.728	0.801	0.617
YOLOX	0.575	0.709	0.627	0.605	0.904	0.684	0.757	0.822	0.726

---

Segmentation Benchmark(MIoU)

Model	2	3	5	11	12	13	14	15	17	MIoU
FCN	0.991	0.780	0.792	0.788	0.957	0.821	0.925	0.872	0.969	0.877
PSPNet	0.990	0.740	0.740	0.777	0.956	0.788	0.925	0.872	0.959	0.861
DeeplabV3+	0.981	0.618	0.558	0.718	0.906	0.760	0.924	0.862	0.936	0.807
DANet	0.991	0.770	0.782	0.767	0.960	0.785	0.927	0.875	0.943	0.867
CGNet	0.985	0.668	0.548	0.729	0.943	0.783	0.932	0.860	0.908	0.817
OCRNet	0.989	0.727	0.642	0.741	0.956	0.756	0.936	0.869	0.959	0.842
BiSeNetV2	0.989	0.946	0.724	0.708	0.953	0.739	0.917	0.862	0.945	0.837

---

Experiment results

---

Setup Environment

For this project, we follow the settings of mmsegmentation/mmdetection.

conda create --name open-mmlab python=3.8 -y
conda activate openmmlab

Find the correct pytorch version of your cuda version from here.

And the mmcv version can be found here.

Setup Datasets

Download the Spacecraft-DS dataset and extract the folder them to data/cityscapes or data/coco. The final folder structure should look like this:

mmsegmentation
├── configs
├── data
│   ├── cityscapes
│   │   ├── leftImg8bit
│   │   │   ├── train
│   │   │   ├── val
│   │   ├── gtFine
│   │   │   ├── train
│   │   │   ├── val
├── ...

mmdetection
├── mmdet
├── tools
├── configs
├── data
│   ├── coco
│   │   ├── annotations
│   │   ├── train2017
│   │   ├── val2017
│   │   ├── test2017
├── ...

Framework Structure

Our project is based on mmsegmentation and mmdetection). Replace the files in mmsegmentation with our project.

The most important files are( spacecraft segmentation ):

mmsegmentation/configs/hrnet/fcn_hr18_4xb2-40k_cityscapes-512x1024.py
mmsegmentation/configs/_base_/datasets/cityscapes.py
mmsegmentation/configs/_base_/models/fcn_hr18.py
mmsegmentation/datasets/cityscapes.py

The most important files are( spacecraft detection ):

mmdetection/configs/hrnet/cascade_rcnn_hrnetv2p_w18_20e_coco.py
mmdetection/configs/_base_/datasets/coco_detection.py
mmdetection/configs/_base_/models/cascade_rcnn_r50_fpn.py
mmdetection/mmdet/datasets/coco.py

Training

bash tools/dist_train.sh configs/hrnet/cascade_rcnn_hrnetv2p_w18_20e_coco.py

Acknowledgements

Our project is based on the following open-source projects.

• MMSegmentation
• MMDetection

Citation

If you find this paper useful in your research, please consider cite:

@Article{
title={Spacecraft-DS: A Spacecraft Dataset for Key Components Detection and Segmentation via Hardware-in-the-loop Capture},
author={Yi Cao, Jinzhen Mu, Xianghong Cheng, and Fengyu Liu},
}