This is an overview of our proposed heterogeneous remote sensing images "SAR2Opt" dataset.
SAR2Opt dataset can be used as a benchmark in change detection and image transaltion on remote sensing images.
We manually selected ground points on each pair of SAR-optical images to perform fine registration. All the pathces were cropped in size of 600*600 pixels after registration. If you want to use a pre-trained model, you could resize the patches first.
You can get the dataset from:
Here are some translated results on our SAR2Opt dataset with well-known GAN-based methods, which have been included in our GRSL paper.
We slightly modofied the original codes to make our dataset suitable for the models. And we are grateful to the authors who have shared their codes kindly.
python train.py --dataroot ./datasets/combined_sar2opt --name sar2opt_pix2pix --model pix2pix --direction AtoB
python test.py --dataroot ./datasets/combined_sar2opt --name sar2opt_pix2pix --model pix2pix --direction AtoB
bash ./scripts/train_scripts/train_combined_sar2opt.sh
bash ./scripts/test_scripts/test_combined_sar2opt.sh
python test.py --dataroot ./datasets/sar2opt --name sar2opt_cyclegan --model cycle_gan
python test.py --dataroot ./datasets/sar2opt --name sar2opt_cyclegan --model cycle_gan
python train.py --config configs/sar2opt_folder.yaml
python test.py --config configs/sar2opt_folder.yaml \
--input_folder datasets/sar2opt/testA/ \
--output_folder results/sar2opt/images/ \
--checkpoint outputs/sar2opt_folder/checkpoint/YOUR_MODEL.pt \
--a2b 1
python main.py --dataset sar2opt
python main.py --dataset sar2opt --phase test
python train.py --dataroot ./datasets/sar2opt --name sar2opt_CUT --CUT_mode CUT
python test.py --dataroot ./datasets/sar2opt --name sar2opt_CUT --model cut
bash my_scripts/train_asgit_v2.sh
bash my_scripts/test_asgit_v2.sh
If you find this dataset valuable in your projects, please cite our paper below:
@article{zhao2022comparative,
title={A Comparative Analysis of GAN-based Methods for SAR-to-Optical Image Translation},
author={Zhao, Yitao and Celik, Turgay and Liu, Nanqing and Li, Heng-Chao},
journal={IEEE Geoscience and Remote Sensing Letters},
year={2022},
publisher={IEEE}
}