Explicit TGV

[paskino]

Issue

closes #1985

Description

Implementation of a Total Generalised Variation regulariser for Least Squares problems. It works both with PDHG and the SPDHG algorithms.

Controls to from the interface that are required:

1. alpha, regularisation parameter as in Eq. 3.5 https://royalsocietypublishing.org/doi/10.1098/rsta.2020.0193
2. gamma, ratio beta/alpha, wrt to Eq. 3.5 https://royalsocietypublishing.org/doi/10.1098/rsta.2020.0193
3. regulariser, whether TV or TGV

Add a description of the changes made.

Testing

Describe the tests that were used to verify your changes.

Acceptance Criteria

How should the reviewer test your changes?

Documentation

How have you changed the documentation to reflect your changes? All changes should be noted in the appropriate file in docs/release_notes

[paskino]

I believe this code is ready to be tested by you @samtygier-stfc @JackEAllen

[JackEAllen]

I believe this code is ready to be tested by you @samtygier-stfc @JackEAllen

Thanks, I'll take a look.

[samtygier-stfc]

@paskino Could you cherry-pick 746e96c faae8eb and bbd84dd from https://github.com/mantidproject/mantidimaging/commits/paskino_explicit_TGV. That should get the tests passing. Thanks

[samtygier-stfc]

Thanks that looks like it solves the non-negative part. The other issues was when stochastic is enabled:

2023-12-05 14:48:19,767 [mantidimaging.gui.dialogs.async_task.task:L53] ERROR: Failed to execute task: tuple index out of range
Traceback (most recent call last):
  File "/home/sam/git/mantidimaging/mantidimaging/gui/dialogs/async_task/task.py", line 50, in run
    self.result = call_with_known_parameters(self.task_function, **self.kwargs)
  File "/home/sam/git/mantidimaging/mantidimaging/core/utility/func_call.py", line 12, in call_with_known_parameters
    return func(**ka)
  File "/home/sam/git/mantidimaging/mantidimaging/gui/windows/recon/model.py", line 144, in run_preview_recon
    recon.data[0] = reconstructor.single_sino(images.sino(slice_idx),
  File "/home/sam/git/mantidimaging/mantidimaging/core/reconstruct/cil_recon.py", line 249, in single_sino
    algo = SPDHG(f=F,
  File "/home/sam/mambaforge/envs/mantidimaging-dev/lib/python3.10/site-packages/cil/optimisation/algorithms/SPDHG.py", line 105, in __init__
    self.set_up(f=f, g=g, operator=operator, tau=tau, sigma=sigma,
  File "/home/sam/mambaforge/envs/mantidimaging-dev/lib/python3.10/site-packages/cil/optimisation/algorithms/SPDHG.py", line 156, in set_up
    norms = [operator.get_item(i,0).norm() for i in range(self.ndual_subsets)]
  File "/home/sam/mambaforge/envs/mantidimaging-dev/lib/python3.10/site-packages/cil/optimisation/algorithms/SPDHG.py", line 156, in <listcomp>
    norms = [operator.get_item(i,0).norm() for i in range(self.ndual_subsets)]
  File "/home/sam/mambaforge/envs/mantidimaging-dev/lib/python3.10/site-packages/cil/optimisation/operators/BlockOperator.py", line 136, in get_item
    return self.operators[index]
IndexError: tuple index out of range

[paskino]

@samtygier-stfc I need to think about how to make SPDHG work. Can it be disabled for now?

[samtygier-stfc]

Thanks. Ok. I'll merge as is, and will disable stochastic for TGV when I do the GUI.

Current test failures look like the applitools API is down, its a secrets issue. so I'll worry about that.

[samtygier-stfc]

closing as we have #1997

[MikeSullivan7]

Looks good, seems to match up with the paper and give good results with a higher number of iterations