Skip to content

Latest commit

 

History

History
61 lines (53 loc) · 3.03 KB

README.md

File metadata and controls

61 lines (53 loc) · 3.03 KB
Raw

UTeRM

Source code and data for the paper
Deep Unfolding Tensor Rank Minimization With Generalized Detail Injection for Pansharpening
Truong Thanh Nhat Mai, Edmund Y. Lam, and Chul Lee
IEEE Transactions on Geoscience and Remote Sensing, vol. 62, pp. 1-18, 2024, Art no. 5405218
https://doi.org/10.1109/TGRS.2024.3392215

For PDF, please visit https://mtntruong.github.io/

Source code

The proposed algorithm is implemented in Python using PyTorch.
We first upload the implementations of the deep unfolded networks and testing codes. Since the networks do not require any special Python library other than Pytorch, you can easily plug them into your training code to train with your own dataset. Please see the toy example in the main function in each file.

Please note that the proposed UTeRM needs data augmentation to achieve the reported performance (we also used the same data augmentation procedure for all deep networks in the experiments to ensure fair comparisons).

Required Python packages

Even though the deep unfolded networks only need Pytorch to run, the training/testing scripts require some external libraries. Please use env.yml to create an environment in Anaconda

conda env create -f env.yml

Then activate the environment

conda activate uterm

If you want to change the environment name, edit the first line of env.yml before creating the environment.

Data Preparation

Please download the raw data from here then run the MATLAB script Data-Preparation/H5_Data_Gen.m to create an H5 file containing preprocessed and augmented data. Please note that all learning-based competing algorithms in the paper use this offline-augmented data for training.

Training

The following script performs the training procedure described in the paper

python train.py --arch UTeRM_CNN --data ./msi_data/IKONOS_train.h5 && python train.py --arch UTeRM_CNN --data ./msi_data/IKONOS_train.h5 --finetune --resume=./ckpts/epoch_90.pth --set_lr=1e-6

The data file ./msi_data/IKONOS_train.h5 serves as an example of the generated H5 file described above.

Testing

Reduced-resolution test

python test_reduced_res.py --arch UTeRM_CNN --data ./msi_data/IKONOS_test.h5 --weight ./checkpoints/UTeRM_CNN.pth

Full-resolution test

python test_full_res.py --arch UTeRM_CNN --data ./msi_data/IKONOS_test.h5 --weight ./checkpoints/UTeRM_CNN.pth

Please note that the file IKONOS_test.h5 only contains a part of the test set used in the paper since GitHub does not allow big files.

Citation

If our research or dataset are useful for your research, please kindly cite our work

@article{Mai2024,
    author={Mai, Truong Thanh Nhat and Lam, Edmund Y. and Lee, Chul},
    journal={IEEE Transactions on Geoscience and Remote Sensing}, 
    title={Deep Unfolding Tensor Rank Minimization With Generalized Detail Injection for Pansharpening}, 
    year={2024},
    volume={62},
    number={},
    pages={1-18},
    doi={10.1109/TGRS.2024.3392215}
}