Remote-sensing-Image-Compression

Remote-sensing Image Compression

This repository contains the code for reproducing the results with trained models (EGA-Net), NWPU-RSC Dataset, and a novel full-reference IQA FITS.

Requirements

Python==3.6
torch==1.6.0
torchvision==0.7.0

Data download

ISPR Vaihingen Data: https://www2.isprs.org/
AID Dataset: https://paperswithcode.com/dataset/aid
NWPU-RSC Dataset: https://pan.baidu.com/s/1XismIJG5iO9aB9hv0VpofQ 提取码：1510

Sample of the NWPU-RSC

Different Visual Tasks on RS Benchmark Dataset

Data Sources of the NWPU-RSC

Visual Results

Experimental results on the AID dataset

Experimental results of different methods on the AID dataset. In each example, the right-most columns represent ground truths for input image. In the other columns, the first row shows the reconstructed results by different methods, and the second row shows the magnified details of the red rectangle.

Experimental results on the NWPU-RSC dataset

Experimental results on the ISPR Vaihingen Dataset

Rate-distortion curves of different compression algorithms on the ISPR Vaihingen Dataset. (a) PSNR, (b) M-SSIM, (c) VIETIM and (d) LPIPS on the ISPR dataset.

Change detection results in different compression methods

Visualization of change detection results.(a) Orginal Images, (b) BPG, (c) JPEG2000, (d) GMM, (e) C2F, (f) EGA-Net (lo), (g) EGA-Net (me) and (h) EGA-Net (hi).

Citation

If you find our project is useful for your research, please cite:

@ARTICLE{10247080,
  author={Han, Pengfei and Zhao, Bin and Li, Xuelong},
  journal={IEEE Transactions on Geoscience and Remote Sensing}, 
  title={Edge-Guided Remote-Sensing Image Compression}, 
  year={2023},
  volume={61},
  number={},
  pages={1-15},
  doi={10.1109/TGRS.2023.3314012}}