Crop training dataset. I randomly select 20 images for training and crop about 10000 patches of size 96*96*20. The corresponding MATLAB codes are in “data/gene_patches.m”. (You could also freely choose your favourite way to crop patches.)
Save the training dataset in path “dataroot”(your own data path).
2.2 Testing dataset
Please refer to “data/gene_test_data.m” file for creating your own test data using MATLAB. The noise generation methods in “data/utils” file are in consistent with those in “lib.py”.
3. Training and testing
Plean refer to “train_hwnet.py” and “test.py” for training and testing HWLRMF. More test codes for NAILRMA, NGmeet, LLRT and their weighted versions will be uploaded soon.
4. Other information
4.1 SVD grad
For pytorch>=1.9, torch.linalg.svd could also be directly used. However, sometimes the grads could be numerically unstable.
5. Citation
If you are interested in our work, please cite
@InProceedings{Rui_2021_CVPR,
author = {Rui, Xiangyu and Cao, Xiangyong and Xie, Qi and Yue, Zongsheng and Zhao, Qian and Meng, Deyu},
title = {Learning an Explicit Weighting Scheme for Adapting Complex HSI Noise},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2021},
pages = {6739-6748}
}
Learning An Explicit Weighting Scheme for Adapting Complex HSI Noise (CVPR2021)
Xiangyu Rui1, Xiangyong Cao1, Qi Xie1, Zongsheng Yue1, Qian Zhao1, Deyu Meng1,2
1Xi’an Jiaotong University; 2Pazhou Lab, Guangzhou
Main paper
Supplement material
1. Basic requirements
python >= 3.8
pytorch = 1.9 (lower version may also be applicable.)
2. Prepare data
2.1 Training dataset
Please download CAVE DATAset from https://www.cs.columbia.edu/CAVE/databases/multispectral/ for training. The image size is of 512*512*31.
Crop training dataset. I randomly select 20 images for training and crop about 10000 patches of size 96*96*20. The corresponding MATLAB codes are in “data/gene_patches.m”. (You could also freely choose your favourite way to crop patches.)
Save the training dataset in path “dataroot”(your own data path).
2.2 Testing dataset
Please refer to “data/gene_test_data.m” file for creating your own test data using MATLAB. The noise generation methods in “data/utils” file are in consistent with those in “lib.py”.
3. Training and testing
Plean refer to “train_hwnet.py” and “test.py” for training and testing HWLRMF. More test codes for NAILRMA, NGmeet, LLRT and their weighted versions will be uploaded soon.
4. Other information
4.1 SVD grad
For pytorch>=1.9, torch.linalg.svd could also be directly used. However, sometimes the grads could be numerically unstable.
5. Citation
If you are interested in our work, please cite
6. Contacts
If you have any questions, please contract me by xyrui@outlook.com or rxy14789653@stu.xjtu.edu.cn.