This page includes an end2end implementation of DeepSpace.
DeepSpace: Super Resolution Powered Efficient and Reliable Satellite Image Data Acquistion
Chuanhao Sun, Yu Zhang (The University of Edinburgh); Bill Tao, Deepak Vasisht (University of Illinois Urbana-Champaign); Mahesh Marina (The University of Edinburgh)
SIGCOMM ’25, September 8–11, 2025, Coimbra, Portugal
We use Anaconda to manage the virtual environment.
conda env create -f environment.yml
conda activate deepspacePlease find the processed showcase dataset on Google Drive
Then unzip the data to .../decompress/data/.
This dataset is based on Planet-California, please cite the original data source as
@article{devaraj2017dove,
title={Dove high speed downlink system},
author={Devaraj, Kiruthika and Kingsbury, Ryan and Ligon, Matt and Breu, Joseph and Vittaldev, Vivek and Klofas, Bryan and Yeon, Patrick and Colton, Kyle},
year={2017}
}The datasets used in this work are open-source, please refer to their original source for further access.
The code can be found in '\BLSH'
Please ensure ImageHash is available before using BLSH module.
Run python detect.py -i [your-target-path] will detet the high similar images in terms of BLSH
Besides the BLSH comparison with reference images, the compression in DeepSpace comes with a simple logic - SSIM based resolution selection.
The corresponding code can be founf in '\Compress\creat_input.py'
You must run BLSH with reference images before running the SSIM-based compress.
The decompress process is based on wavelet diffusion, where the SR model for each scenario and resolution shall be trained separately.
To train the model, simply use the ..\decompress\run.sh
Similarly, ..\decompress\test_srwddgan.py specifies how to evaluate a trained model.