Copyright

All rights reserved. This work should only be used for nonprofit purposes.

Description

Packages for a benchmarking protocol for SAR colorization which contains a protocol for generating ground truth, data analysis algorithm, three spectral-based solutions, three deep-learning-based solutions, metrics, and performance inspections.

Protocol

• "filepath_gen.py" is for generating the file path of the dataset. The last null string should be deleted manually.
• "gt_ihs_gen.m" is for generating ground truth images through the GIHS algorithm.

Spectral_based_solutions

• codes for three spectral-based solutions, namely NoColSAR, LR4ColSAR, and NL4ColSAR.
• there is also a code for data analysis through scatter plots between the SAR band and the band of ground truth.

CNN4ColSAR

• codes for spatial-spectral-based solution CNN4ColSAR.
• a test model file is also contained.

DivColSAR

• codes for solution DivColSAR.

cGAN4ColSAR:

• codes for solution cGAN4ColSAR.

Models

• saved models of the proposed approaches for reproducibility.
• link: https://pan.baidu.com/s/1yw9BtVMB4_WQcZwPMmzhRQ?pwd=cwmd, password: cwmd. You can also use the link: https://terabox.com/s/1OLJyYWKOtO66eT7cw42Njw

Metrics

• Matlab codes for Q2n.
• python codes for SAM and NRMSE.

Performance_inspection

• Matlab code for data analysis between colorized SAR image and ground truth.
• python code for residual image comparison.

SEN12MS_CR_SARColorData

• there are 4 train samples and 4 test samples.
• the whole train dataset can be downloaded by the following Baidu Netdisk link: https://pan.baidu.com/s/1xHSDNXoQzo5xewMsjQCBrA?pwd=2h7y, password：2h7y. You can also use the link: https://terabox.com/s/1J2mTvAROV9SSzilUkwsgeg
• it is noted that the sar_train and opt_train are divided into several parts because of some restrictions, so the users can merge them after decompression. Additionally, sar_train_20, opt_train_20, and gt_train_20 are the trainset for LR4ColSAR and NL4ColSAR.

Citation

If you find this code helpful, please kindly cite our paper.
@article{SHEN2024698,
title = {A benchmarking protocol for SAR colorization: From regression to deep learning approaches},
journal = {Neural Networks},
volume = {169},
pages = {698-712},
year = {2024},
author = {Kangqing Shen and Gemine Vivone and Xiaoyuan Yang and Simone Lolli and Michael Schmitt}
}