tomographicimaging / cil Goto Github PK

• parallel 3D
• cone 2D
• cone 3D

• CGLS code
  • parallel beam
  • cone beam

A class hierarchy will be nice to organise similar algorithms together, and common methods and attributes could be inherited from superclass(es).

Should be able to restart the CGLS from the last iteration.

• SB_TV
• TGV
• Patch based
• Diffusion
• LLT

DataSet should be able to save to HDF5

Build on top of DataSetProcessor

Test it

Create dataset for multichannel data.

Add dimension to the 2/3D DataSets (volume/sinogram) and add energy dimension.

• ASTRA
  • parallel beam
  • cone beam

As we have many similarities with the CCPPETMR reconstruction framework SIRF (and shared developers).

Let us align more on the naming conventions: see SIRF Developer Guide.

https://github.com/CCPPETMR/SIRF/blob/master/doc/UserGuide.md#basic-classes-

Volume Geometry object

It may be useful to have a DataSetProcessor to calculate the average of a DataSet along an axis and collapse that axis. For instance, when flats and darks are calculated they are a stack of N images.
It may be useful to have a processor that averages the slices out and creates a 2D DataSet out of it.

• ROF_TV
• FGP_TV

DataSet loader.
Returns DataSet (SinogramData) and Geometry

When going public, if not including support for importing FGP_TV etc from the FISTA/regularizer repo, I suppose we should move the demo_IP from the master branch to a separate (experimental) branch?

all or any

Finish the DataSetProcessor

Use the code from instrument

• Nexus
• Xtek
• XRadia
• Tiff stacks

Move geoms to framework.

• Diamond single channel parallel beam dataset
• SophiaBeads single channel
• Demo IP (multichannel)

These are needed for Norm1 func and were present in 56bef36

SinogramGeometry object

Arithmetic operations etc. explicitly (hard coded) produce generic DataSet objects, despite original objects being from specific subclass. This for example means that after running FISTA, the reconstruction is a DataSet and not, as desired, a VolumeData. Perhaps possible to fix by replacing hard coded "DataSet" by "type(self)".

This is complementary to the subset method extracting a single-channel data set from a specific channel of a multi-channel data set.

• FGP
• ROF

This is useful to remove zingers

Data loader.

Return SinogramData and Geometry

All the algorithms do iterate up to the specified number of iterations.
there is no stopping rule.

Running simple_demo in 59c01d8 causes the error

Traceback (most recent call last):

  File "<ipython-input-10-cd664ec232c5>", line 1, in <module>
    x_CGLS, it_CGLS, timing_CGLS, criter_CGLS = CGLS(Aop, b, 1000, x_init)

  File "..\ccpi\reconstruction\algs.py", line 142, in CGLS
    d = A.adjoint(r)

  File "..\ccpi\reconstruction\astra_ops.py", line 52, in adjoint
    self.bp.setInput(DATA)

  File "..\ccpi\framework.py", line 628, in setInput
    if self.checkInput(dataset):

  File "..\ccpi\astra\astra_processors.py", line 128, in checkInput
    if dataset.number_of_dimensions == 3 or dataset.number_of_dimensions == 2:

AttributeError: 'SinogramData' object has no attribute 'number_of_dimensions'

• Volume Geometry
• Sinogram Geometry (experimental geometry)

• Do something
• Do Something else