Deep Compressive Sensing in Caffe

December 10, 2018 ยท View on GitHub

Collection of source code fore deep learning based compressive sensing in Caffe.

Disclaimer

All tools are actually reimplemented with some improvements as

  1. For ReconNet and DR2Net:
  • Implemented as fully-convolutional network
  • Generated files with pycaffe (train, test, solver - prototxt as well as train.sh).
  • Data generation is added: sampling matrices, separate training data for different subrate, more test data, etc.
  • General test framework: applied for both ReconNet and DR2Net

Collection Tools

  1. ReconNet https://github.com/KuldeepKulkarni/ReconNet
  2. DR2Net https://github.com/AtenaKid/caffe_dr2

References

[1] K. Kulkarni et al,"ReconNet: Non-Iterative Reconstruction of Images From Compressively Sensed Measurements", CVPR 2016.

[2] H. Yao et al, "DR2-Net: Deep Residual Reconstruction Network for Image Compressive Sensing", arxiv, 2017