• 设为首页
  • 点击收藏
  • 手机版
    手机扫一扫访问
    迪恩网络手机版
  • 关注官方公众号
    微信扫一扫关注
    迪恩网络公众号

yangyan92/ADMM-CSNet: Matlab code for ADMM-CSNet

原作者: [db:作者] 来自: 网络 收藏 邀请

开源软件名称(OpenSource Name):

yangyan92/ADMM-CSNet

开源软件地址(OpenSource Url):

https://github.com/yangyan92/ADMM-CSNet

开源编程语言(OpenSource Language):

Cuda 46.2%

开源软件介绍(OpenSource Introduction):

ADMM-CSNet

Generic-ADMM-CSNet


These are testing and training codes for Generic-ADMM-CSNet in "ADMM-CSNet: A Deep Learning Approach for Image Compressive Sensing" (TPAMI 2019)

If you use thses codes, please cite our paper:

[1] Yan Yang, Jian Sun, Huibin Li, Zongben Xu. ADMM-CSNet: A Deep Learning Approach for Image Compressive Sensing (TPAMI 2019).

http://gr.xjtu.edu.cn/web/jiansun/publications

All rights are reserved by the authors.

Yan Yang -2019/04/10. For more detail or traning data, feel free to contact: [email protected]


Data link:

https://pan.baidu.com/s/1nvf07g_OmMAnFAbhG1orIQ passwards:sdsq

Usage:

  1. Three folders.

    1). 'Generic-ADMM-CSNet-ComplexMRI' are testing and training codes to reconstruct complex-valued MR images with 1D Cartesian masks and 2D random masks.
    2). 'Generic-ADMM-CSNet-RealMRI' are testing and training codes to reconstruct real-valued MR images with the Pseudo radial mask.
    3). 'Generic-ADMM-CSNet-Image' are testing and training codes to reconstruct natural images with the randomly permuted coded diffraction operators and Walsh-Hadamard operators.

    Please do not add these three folders into the path at the same time, because they contain the functions with the same name.

  2. For testing the trained network for a single image.
    ('./Generic-ADMM-CSNet-ComplexMRI/main_ADMM_CSNet_test.m')
    ('./Generic-ADMM-CSNet-RealMRI/main_ADMM_CSNet_test.m')
    ('./Generic-ADMM-CSNet-Image/main_ADMM_CSNet_test.m')

    1). Load trained network with different stages in main_ADMM_CSNet_test.m.
    If you apply ADMM-CSNet to reconstruct other MR or natural images, it is best to re-train the models.

     E.g., The model './net/NET-1D-Cartesian-0.2-complex-S10.mat' is the ADMM-CSNet with 10 stages trained from 100 complex-valued MR images using 1D Cartesian mask with 20% sampling rate.
           The model './net/NET-Pseudo-radial-0.2-real-S11.mat' is the network with 11 stages trained from 100 real-valued MR images using Pseudo radial mask with 20% sampling rate.
           The model './net/Net-Diffraction-0.05-S10.mat' is the network with 10 stages trained from 100 real-valued natural images using coded diffraction operator with 5% sampling rate.
    

    2). Load test image in main_ADMM_CSNet_test.m
    The images in './data/Brain_complex_data', './data/Brain_real_data', './data/Image' are fully-sampled images.

    3). Load sampling mask or operator with different sampling ratios in main_ADMM_CSNet_test.m

     E.g., The mask './mask/1D-Cartesian-0.2.mat' is a 1D Cartesian mask with 20% sampling rate.
           The mask './mask/D-0.1.mat' is a coded diffraction operator with 10% sampling rate. 
    

    4). Network testing setting (network structure or training setting) is in 'config.m '.

    5). To test our ADMM-CSNet, run 'main_ADMM_CSNet_test.m'

  3. For testing the trained network for our testing dataset.
    ('./Generic-ADMM-CSNet-ComplexMRI/AverageTesting.m')
    ('./Generic-ADMM-CSNet-RealMRI/AverageTesting.m')
    ('./Generic-ADMM-CSNet-Image/AverageTesting.m')

    1). Load trained network with different stages in AverageTesting.m.
    If you apply ADMM-CSNet to reconstruct other MR or natural images, it is best to re-train the models.

     E.g., The model './net/NET-1D-Cartesian-0.2-complex-S10.mat' is the ADMM-CSNet with 10 stages trained from 100 complex-valued MR images using 1D Cartesian mask with 20% sampling rate.
           The model './net/NET-Pseudo-radial-0.2-real-S11.mat' is the network with 11 stages trained from 100 real-valued MR images using Pseudo radial mask with 20% sampling rate.
           The model './net/Net-Diffraction-0.05-S10.mat' is the network with 10 stages trained from 100 real-valued natural images using coded diffraction operator with 5% sampling rate.
    

    2). Set the data_dir of testing dataset ans load the correspongding mask in AverageTesting.m

     E.g., data_dir = './data/DATA-1D-Cartesian-0.2-complex-brain/test/' is the testing dataset including 100 complex-valued brain MR image with 20% 1D-Cartesian mask.  
           data_dir = './data/Testingdata/Sdata10/D/D_0.1_1/' is the first testing dataset including 10 standard image with 10% coded diffraction operator.  
           data_dir = './data/DATA-Pseudo-radial-0.2-real-brain/test/'is the testing dataset including 50 real-valued brain MR image with 20% Pseudo radial mask.  
    

    3). Network testing setting (network structure or training setting) is in 'config.m '.

    4). To test our ADMM-CSNet, run 'AverageTesting.m'

  4. For re-training the ADMM-CSNets

    1). Set the data_dir of training dataset ans load the correspongding mask in L_BFGSnetTrain.m.

    2). Modify the network setting and trainging setting in 'config.m '.

    3). To train ADMM-CSNet by L-BFGS algorithm, run ' L_BFGSnetTrain.m' .

    4). After training, the trained network and the training error are saved in './Train_output'.





鲜花

握手

雷人

路过

鸡蛋
该文章已有0人参与评论

请发表评论

全部评论

专题导读
热门推荐
阅读排行榜

扫描微信二维码

查看手机版网站

随时了解更新最新资讯

139-2527-9053

在线客服(服务时间 9:00~18:00)

在线QQ客服
地址:深圳市南山区西丽大学城创智工业园
电邮:jeky_zhao#qq.com
移动电话:139-2527-9053

Powered by 互联科技 X3.4© 2001-2213 极客世界.|Sitemap