Everything comes from the code by Zachary Harmany ([email protected]).

Original source: http://drz.ac/code/spiraltap/ or https://github.com/zharmany/SPIRAL-TAP (might be more up-to-date). The algorithm is described in this paper:

Z. T. Harmany, R. F. Marcia, and R. M. Willett, "This is SPIRAL-TAP: Sparse Poisson Intensity Reconstruction ALgorithms – Theory and Practice," IEEE Transactions on Image Processing, vol. 21, pp. 1084–1096, Mar. 2012.

If you use this code for academic work, please mention it:

2. Code: still experimental.

pip install pySPIRALTAP

pySPIRALTAP requires the following dependencies:

• numpy
• rwt, the Rice Wavelet Toolbox, available on this page: https://github.com/ricedsp/rwt
• scipy.io (only to run the demo)
• pytest (only to run the tests)

Installation using your package manager (Debian/Ubuntu):

sudo apt-get install python-numpy

Alternatively, if you have pip installed, you can install numpy by typing the following:

pip install --user numpy

You can install the Rice Wavelet Toolbox by typing the following. Before that, make sure that you have cmake installed (sudo apt-get install cmake if you use a Debian-derived distribution):

git clone https://github.com/ricedsp/rwt.git
cd rwt/python
cmake .
sudo make install

And then test if the installation succeeded by typing:

python -c "import rwt"

If this returns nothing, the installation worked. In case it returns an error such as:

Traceback (most recent call last):
  File "<string>", line 1, in <module>
ImportError: No module named rwt

This means that the installation failed. It can be because the installation was performed using a different version of Python that is not the default version. So far, I do not have information about how to perform this installation.

More information about the installation can be found in the INSTALL file, or on the rwt webpage.

Installation can be performed using the following shell commands:

git clone https://gitlab.com/padouppadoup/pySPIRAL-TAP.git
cd pySPIRAL-TAP
sudo python setup.py install

A simple working example can be run at (in the main folder):

python SPIRALdemo.py

This file contains one demo where a 1D signal is reconstructed using a $l1$ penalty. This demo uses matplotlib for plotting output.

Alternatively, one can play with the Jupyter notebook: SPIRALdemo.ipynb, that feature the same demo as SPIRALdemo.py, but in a more fancy format.

The pySPIRALTAP methods can be imported with import pySPIRALTAP.

Here is a canonical function call with many parameters exposed:

    resSPIRAL = pySPIRALTAP.SPIRALTAP(y,A,              # y: measured signal, A: projection matrix
		                              1e-6,             # regularization parameter
                                      maxiter=100,      # min. number of iterations
                                      miniter=5,        # max. number of iterations
                                      penalty='canonical', # type of penalty to apply
                                      noisetype='gaussian',# form of the log-likelihood penalty
                                      stopcriterion=3,  # index of the termination criterion
                                      tolerance=1e-8,
                                      alphainit=1,
                                      alphamin=1e-30,
                                      alphamax=1e30,
                                      alphaaccept=1e30,
                                      logepsilon=1e-10,
                                      saveobjective=True,
                                      savereconerror=True,
                                      savesolutionpath=False,
                                      verbose=verbose, savecputime=True)

The methods based on RDP (recursive dyadic partitions) have not been implemented yet. Additionnally, the code has not been fully tested, although we did our best to provide a working product.

• This software is released under the MIT license. See the LICENSE file for more details.
• The denoise_bound code is released under the GNU GPLv2 license and was written by Copyright (2008): Amir Beck and Marc Teboulle.