LapCSNet

• Officical code of paper "An efficient deep convolutional laplacian pyramid architecture for CS reconstruction at low sampling ratios" ICASSP2018
• Download the paper: https://arxiv.org/pdf/1804.04970.pdf

Framework of LapCSNet

Requirements

• Windows10
• Matlab R2015b
• MatconvNet 1.0-beta23 (https://www.vlfeat.org/matconvnet/)
• CUDA 8.0

How to Run

Training

• Copying the function in +dagnn folder to your Matconvnet location <MatconvNet>\matlab\+dagnn
• Preparing the training data. (T91 and BSDS200 are included in our repo)
• Train the LapCSNet, run the code train_LapCSN(0.1, 2, 0);

The first param is CS subrate
The second param is the number of conv layers in each pyramid level
The third param is gpu setting. (0 is CPU, 1 is GPU)

Testing

• Preparing the testing data. (Set5 and Set14 are included in our repo)
• Test the LapCSNet, run the code test_LapCSN_main(100, 200)

The first param is start epoch for testing model
The second param is end epoch for testing model 

Experimental Results

• Subjective results

• Objective results

Additional instructions

• For training data, you can choose any dataset by yourself.
• When subrate<=0.25, the laplacian structure can be used.
• If you like this repo, Star or Fork to support my work. Thank you.
• If you have any problem, please email wxcui@hit.edu.cn

Citation

• If you find the code is useful in your research, please cite:

@article{Cui2018An,
  title={An efficient deep convolutional laplacian pyramid architecture for CS reconstruction at low sampling ratios},
  journal={IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
  author={Cui, Wenxue and Xu, Heyao and Gao, Xinwei and Zhang, Shengping and Jiang, Feng and Zhao, Debin},
  year={2018},
}

Acknowledgments

This code is built based on the repo https://github.com/phoenix104104/LapSRN