[SyneRBI/SIRF-Contribs] MaGeZ (PR #28)

Casper da Costa-Luis

unread,

Jan 23, 2026, 9:25:40 AMJan 23

to SyneRBI/SIRF-Contribs, Subscribed

merge SyneRBI/PETRIC-MaGeZ@65548b7
zerocopy tweaks to reduce memory usage

/CC @gschramm

You can view, comment on, or merge this pull request online at:

https://github.com/SyneRBI/SIRF-Contribs/pull/28

Commit Summary

3f0e6b9 add optional argument for giving image size
2d18cb9 preliminary version of BSREM for the mMR NEMA IQ
fff935a some more early checks
18abec4 preliminary version of QC utility
871346b make Siemens mMR data appear in better directories
bd28698 reduce duration of mMR data from 600 to 500
8c8a55e Merge branch 'main' into BSREM_scripts
f1f67d6 more plots
2fb23d2 2 extra BSREM scripts
a54540b fixed filename
f1434ea parameterise BSREM recon
101e810 lint
c809414 add tensorboardX
32ba086 write images and all objectives
21bc466 fix bugs
72d4cec drop matplotlib, log all intervals, add clip & cmax
ab32faa improve BSREM_common.py
2d0a911 separate SaveCallback
2593c5e Merge branch 'BSREM_scripts' of github.com:SyneRBI/PETRIC into BSREM_scripts
b6ac6f3 add optional savedir to SaveCallback
d23ca8f updated README
d60a895 read penalty weight from file (if it exists)
e88faaa update BSREM files for data-sets
aa10f2a misc tidy
e65a665 lint
a82141d Merge pull request #16 from SyneRBI/BSREM_scripts
80efdd6 main (submission) & petric (backend) framework
2e98031 CI: fix setup
81cab86 update readme
53b1946 minor comment
7ba4b34 optional subsets & priors
2cca32a drop more Dataset items
64d9afc some docs
b4e3d85 docker: edge image
a8d3155 Merge pull request #20 from SyneRBI/main-backend
cc5448e Update README.md
36156a1 Merge pull request #21 from SyneRBI/update-example-description
0160194 slight docs tweaks
0e472e1 Working SGD example with metrics and SIRF objective functions
e35b5bf add SGD example
67ce652 update deps
9d8e988 Merge pull request #23 from SyneRBI/sgd-example
c371fda image quality callback
5aec470 improve local run docs
ac6401c some review comments
5483ba6 Merge pull request #25 from SyneRBI/metrics-and-docs
01178df document --gpus all
0c26d84 fix potentially missing petric module
fd7cd97 continue on error
aed5cf8 enable logging
5875054 fix per-dataset timeout
5ab9d79 added non-negativity
14e3dc3 git ignore output/ and other files
68756d4 import petric no longer reads data
ee948ef typos, and re-instate setting first dataset
99266c6 petric: get_data lazy load (#32)
a78fb4a fix checks
c607fcf IndicatorBox numpy implementation
fcd5f71 fiddle with parameters
a4b75e7 final cleanup
e3f5300 Merge pull request #35 from SyneRBI/ISTA
ba74194 use OSEM image
2bd78f4 added preconditioner
dc6a478 add kappa to data
5b1431f lint + print exceptions
fffc8bd Apply suggestions from code review
970bb18 lint & comments
4301b7c Merge pull request #36 from SyneRBI/precond
46ec6b6 note about renaming/symlinking
c45b39b remove redundant exception types
7fa22f8 update ISTA step_size
85393b1 Merge pull request #39 from SyneRBI/ISTA_step_size
b043a13 add metrics as per wiki
d397e0e update after review
d7e6fe4 drop filters
a5860b5 slight tidy
beb42e9 Merge pull request #46 from SyneRBI/metrics
479a16b add QC for reference image and VOIs
244d715 pass through options and make saving optional
35e45aa Merge pull request #48 from SyneRBI/data_QC_VOIs
9900487 fix metrics
bc5d6c5 more options for data_QC (#52)
23c396a Use CUDA RDP if it exists (#49)
fa30ba7 fix objectives.csv
f50765f lint
c6994b3 Merge pull request #53 from SyneRBI/objectives-csv
086f029 slight doc updates
f015da8 replace namedtuple => dataclass
dc556ce drop typing.Any
4723355 Merge pull request #55 from SyneRBI/dataclass
247df6c fix mutable default error
6b48deb provide example implementing OSEM
c8146b5 misc tidy
e56e20d update docstring
835f555 update docstring
a62efb0 Merge pull request #54 from SyneRBI/OSEM_example
56e36e8 drop symlink
35a6c4c Merge pull request #56 from SyneRBI/drop-symlink
facebd1 metrics: standalone `log()`
f31d077 QualityMetrics.evaluate
11112d2 Merge pull request #57 from SyneRBI/metrics-log
2ace675 SaveIters: flush csv each row
78d6904 Merge pull request #60 from SyneRBI/csv-flush
96b66c7 callbacks with custom interval
e21ff31 data_QC: plot VOIs with transparancy as well
2a741e9 data_QC: centre VOI images and add COM to plots
c5d554c data_QC: fix location of VOI*.png
a9ac16c Merge pull request #61 from SyneRBI/data_QC_VOI
d1abbc7 Merge pull request #59 from SyneRBI/metrics-interval
4a47f7b minor correction in header
640a3fc add some utilities for reading/plotting
f6ff8a9 update README
966d01f lint
67781a4 readability updates
9cc4a52 more pythonic implementation to get keys()
6a0c66b Merge pull request #65 from SyneRBI/evaluation_utilities
4c46a54 tensorboard walltime excluding PETRIC callbacks
00771cf minor correction in header
0df0224 update BSREM ref generation scripts
c4ae648 OSEM check Vision600_thorax
10dda2b plot BSREM metrics
0d6800b OSEM NeuroLF_Hoffman and mMR_NEMA_IQ check-in
7d4b113 add Siemens_Vision600_thorax_VOI_prep.sh
3730147 add NeuroLF_VOIs.py
4fbd847 add Siemens_mMR_NEMA_VOIs.py
6b1e738 add __init.py__
e92046a avoid unnecessary `time()` call
d38a919 Merge pull request #67 from SyneRBI/tb-walltime-offset
d1044ff pass_index
7f3cb70 pass_index(window)
8256725 use scipy.ndimage.binary_erosion
d9fbdc0 Merge pull request #68 from SyneRBI/metrics-pass
27f9e2a raise runtime to 10 mins
daa20bd Merge pull request #75 from SyneRBI/bump-runtime
16f2b64 add examples for common pip/conda packages
62fcfb0 preliminary utility for evolution of metrics
085b2d0 no longer require a template image for the initial OSEM
2bdf0ba Allow initial data preparation without mu-map
e9f917c first set of files for the mMR ACR data
4554816 lint
3fa3bea Merge pull request #78 from SyneRBI/BSREM_metric_plots
fb2658e add note on conda vs pip
e6adc67 Merge pull request #76 from SyneRBI/example-conda-pip
c40e87d merging main...
4f2dbd1 a couple of small corrections
89f0bca [ci skip] described EO corrections in Siemens_mMR_ACR/README.md
b1dd679 misc tidy & import fixes
e72d5ca expose data.path
730325b Merge pull request #80 from SyneRBI/data-path
dd25d08 slight lint
2de2769 docker dev instructions
eef8f6c Merge pull request #81 from SyneRBI/docker-user-run
94c8022 Merge branch 'main' into Siemens_mMR_ACR
181f864 docs: leaderboard config & metrics runtime
5b584f6 Merge pull request #83 from SyneRBI/doc-leaderboard-metrics
8cea939 docs: slight tidy
14ee30a removed comments explaining EO fixes
50228d6 Merge pull request #79 from SyneRBI/Siemens_mMR_ACR
d65095b linting
0777291 mention wiki FAQs
7d26561 update Siemens_mMR_ACR README with additional steps
df947d9 initial BSREM for Siemens_mMR_ACR
6641dc9 Merge branch 'main' into kris_eval
d6cece0 white-space
72f0de4 moved data-set specific files to subfolders
7109574 reduce overlap in evaluation scripts
52c491c some corrections in VOI conversion scripts
0755f0a Merge remote-tracking branch 'origin/main' into kris_eval
ea9ad20 updates to README
9fc66ff remove some user-specific paths
48876af use correct name for folder of NeuroLF data
cc8da8b Merge branch 'kris_eval' into Siemens_mMR_ACR2
1940302 clarify path of template_image
cc83b45 add the_orgdata_path for downloaded data etc
ac3531f add Siemens_mMR_ACR
260a847 Siemens_mMR_ACR processing updates
c9163a5 lint
ced6a57 import & module tidy
c4819ab remove remnant VOI_checks (now in data_QC)
513473a Merge pull request #89 from KrisThielemans/kris_eval
92d40a7 add VOIs for the mMR ACR
6fdda40 Merge remote-tracking branch 'origin/main' into Siemens_mMR_ACR2
cba1219 removed obsolete file
d9d249f Merge pull request #88 from SyneRBI/Siemens_mMR_ACR2
e114f21 lint
0c9cd2c fix prior
fb14c87 Merge pull request #92 from SyneRBI/fix-ISTA-prior
24f69b5 fix conda environment.yml
50f28f0 fix conda environment.yml
c2b4d38 Merge remote-tracking branch 'upstream/main'
192a757 allow dataset sagittal_slices setting
7ee18b7 data_QC improvements
4f1e1cf add Mediso_NEMA_IQ
9097ae2 restore running 400 OSEM updates in the evaluation scripts
281d57e small fixes to evualuation script
a142349 print VOI volume in data_QC
2b8f10c some extra comments in Mediso_NEMA_IQ/README
585f580 Merge pull request #102 from KrisThielemans/Mediso_NEMA_IQ
4bda727 slight lint
b1e31da TensorBoard: add sagittal slice
5c1fa8a Merge pull request #104 from SyneRBI/sagittal
955bd0e add DATA_SLICES
4a8a865 Merge pull request #105 from SyneRBI/data-slices
12f1530 update instructions for creating a new dataset
8eab859 correct default for VOI file
d89e348 git ignore some more output files
f867014 Use dataset_name to get default path
be0c011 use more Python modules
1a16838 Merge pull request #107 from SyneRBI/prep_README
aae6ed3 Threshold before computing sqrt for kappa
cb70ded added Siemens_mMR_NEMA_IQ_lowcounts
375a537 added README
66581cc correct hyperlinks etc
c6719d1 Update petric.py
49ac42e Merge pull request #108 from SyneRBI/Siemens_mMR_NEMA_IQ_lowcounts
a6a5002 Update README.md
0bb809a Create LICENSE
08e6ff0 install pytorch-cuda & tensorflow-gpu
00bc800 Merge pull request #112 from SyneRBI/gpu-libs
dcc50a6 addition of the GE_DMI3_Torso dataset
8bf4314 Merge pull request #111 from SyneRBI/GE_DMI3_Torso
4e8164f add mMR_ACR
f351359 add Mediso_NEMA
069f462 fine-grained metrics, skip objective
73490ac add GE_DMI3_Torso
23a1699 add mMR_NEMA_lowcounts
429d8f6 compute metric every 3 updates for now
e549237 Merge pull request #91 from SyneRBI/mMR_ACR
0eb98be add FOV_mask to Dataset
8e8a7c4 Merge pull request #116 from SyneRBI/FOV_mask
a4915ea update gitignore
0f704b9 resolve merge conflicts
f25f2b1 update README
6096dbc add ALG1 files
7fde2c2 update README
06e67a0 use provided FOV mask
d035b98 remove MaxIteration callback
a196822 code cleanup + documentation
2effc18 correct alg1 description in README
cec28de add ALG2 desctription to README
19132d0 Update README.md
e56220c merge of paper branch that contains new simulations (#3)
716acd9 Update README.md (#4)
2921277 fix use of to_device (#6)
a690205 sim_utils to_device fix (#7)
5890cd0 Cupy fallback (#9)
65548b7 update RDP (boundary conditions) such that they implementation is the exact same (up to float precision) with the one used in STIR's cuda relative difference prior (#11)
4fe14a5 few memory optimisations
9c2a2ef tidy for merge into SIRF-Contribs
0d5ff91 add MaGeZ

File Changes

(5 files)

A src/Python/sirf/contrib/MaGeZ/LICENSE (201)
A src/Python/sirf/contrib/MaGeZ/README.md (170)
A src/Python/sirf/contrib/MaGeZ/environment.yml (6)
A src/Python/sirf/contrib/MaGeZ/main.py (390)
A src/Python/sirf/contrib/MaGeZ/rdp.py (367)

Patch Links:

—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you are subscribed to this thread.

Casper da Costa-Luis

unread,

Jan 23, 2026, 9:29:48 AMJan 23

to SyneRBI/SIRF-Contribs, Push

@casperdcl pushed 2 commits.

1728e54 tidy for merge into SIRF-Contribs
289b7e6 add MaGeZ

—
View it on GitHub or unsubscribe.
You are receiving this because you are subscribed to this thread.

Casper da Costa-Luis

unread,

Jan 23, 2026, 9:34:47 AMJan 23

to SyneRBI/SIRF-Contribs, Push

@casperdcl pushed 2 commits.

a9a0019 tidy for merge into SIRF-Contribs
21ca8e3 add MaGeZ

—
View it on GitHub or unsubscribe.
You are receiving this because you are subscribed to this thread.

Casper da Costa-Luis

unread,

Jan 23, 2026, 9:44:21 AMJan 23

to SyneRBI/SIRF-Contribs, Subscribed

@casperdcl commented on this pull request.

In src/Python/sirf/contrib/MaGeZ/__init__.py:

> @@ -0,0 +1 @@
+from .main import Submission, submission_callbacks

potentially want

⬇️ Suggested change

-from .main import Submission, submission_callbacks
+from .main import Submission as MaGeZ, submission_callbacks

—
Reply to this email directly, view it on GitHub, or unsubscribe.

You are receiving this because you are subscribed to this thread.

Casper da Costa-Luis

unread,

Jan 23, 2026, 9:46:45 AMJan 23

to SyneRBI/SIRF-Contribs, Push

@casperdcl pushed 2 commits.

53e81a7 tidy for merge into SIRF-Contribs
e84322c add MaGeZ

—
View it on GitHub or unsubscribe.
You are receiving this because you are subscribed to this thread.

Casper da Costa-Luis

unread,

Jan 23, 2026, 9:50:31 AMJan 23

to SyneRBI/SIRF-Contribs, Push

@casperdcl pushed 2 commits.

a9cd357 tidy for merge into SIRF-Contribs
3ab9ae3 add MaGeZ

—
View it on GitHub or unsubscribe.
You are receiving this because you are subscribed to this thread.

Kris Thielemans

unread,

Jan 23, 2026, 12:36:00 PMJan 23

to SyneRBI/SIRF-Contribs, Subscribed

@KrisThielemans commented on this pull request.

In src/Python/sirf/contrib/MaGeZ/__init__.py:

> @@ -0,0 +1 @@
+from .main import Submission, submission_callbacks

good idea

—
Reply to this email directly, view it on GitHub, or unsubscribe.

You are receiving this because you are subscribed to this thread.

Kris Thielemans

unread,

Jan 23, 2026, 12:42:42 PMJan 23

to SyneRBI/SIRF-Contribs, Subscribed

@KrisThielemans requested changes on this pull request.

some minor house-keeping suggestions, but I'm a little bit reluctant to add 75 commits here.

Potentially @casperdcl squashes the first PETRIC commits into 1, and then the actual MaGeZ commits (authored by Georg), and then keep the "clean-up" ones separate?

Maybe too much effort fo rnot enough benefit.

In src/Python/sirf/contrib/MaGeZ/LICENSE:

> +      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]

@gschramm can you suggest what this should be?

In src/Python/sirf/contrib/MaGeZ/main.py:

> @@ -0,0 +1,388 @@
+"""Main file to modify for submissions.
+
+Once renamed or symlinked as `main.py`, it will be used by `petric.py` as follows:

probably needs to change

In src/Python/sirf/contrib/MaGeZ/main.py:

> +    The organisers try to `Submission(data).run(inf)` i.e. for infinite iterations (until timeout).
+    This callback forces stopping after `max_iteration` instead.
+    """
+
+    def __init__(self, max_iteration: int, verbose: int = 1):
+        super().__init__(verbose)
+        self.max_iteration = max_iteration
+
+    def __call__(self, algorithm: Algorithm):
+        if algorithm.iteration >= self.max_iteration:
+            raise StopIteration
+
+
+class Submission(Algorithm):
+    """
+    Better preconditioned SVRG for PETRIC (MaGeZ ALG1)

should this therefore be called MaGeZ1?

In src/Python/sirf/contrib/MaGeZ/README.md:

> +### ALG2 (ALG2 branch, ALG2 tag)
+
+Similar to ALG1 but:
+The stepsize is set to 2.2 in the first epoch, and is then computed using the Barzilai-Borwein rule as described in Algorithm 2 in [https://arxiv.org/abs/1605.04131](https://arxiv.org/abs/1605.04131), 
+with m = 1 but using the short stepsize (see [https://en.wikipedia.org/wiki/Barzilai-Borwein_method]([https://en.wikipedia.org/wiki/Barzilai-Borwein_method])) adapted to preconditioned gradient ascent methods
+
+### ALG3 (ALG3 branch, ALG3 tag)
+
+Using the same setup as in ALG2 but with two minor differences.
+First, we use a slightly smaller number of subsets.
+Second, we use a non-stochastic selection of subset indices by considering the cyclic group corresponding to the given number of subsets, finding all of its generators (i.e. the set of all coprimes of the number of subsets that are smaller), and then creating indices by consider specific generators at a time.

not in this PR?

In src/Python/sirf/contrib/MaGeZ/README.md:

> +
+Similar to ALG1 but:
+The stepsize is set to 2.2 in the first epoch, and is then computed using the Barzilai-Borwein rule as described in Algorithm 2 in [https://arxiv.org/abs/1605.04131](https://arxiv.org/abs/1605.04131), 
+with m = 1 but using the short stepsize (see [https://en.wikipedia.org/wiki/Barzilai-Borwein_method]([https://en.wikipedia.org/wiki/Barzilai-Borwein_method])) adapted to preconditioned gradient ascent methods
+
+### ALG3 (ALG3 branch, ALG3 tag)
+
+Using the same setup as in ALG2 but with two minor differences.
+First, we use a slightly smaller number of subsets.
+Second, we use a non-stochastic selection of subset indices by considering the cyclic group corresponding to the given number of subsets, finding all of its generators (i.e. the set of all coprimes of the number of subsets that are smaller), and then creating indices by consider specific generators at a time.
+
+---
+---
+---
+
+# PETRIC: PET Rapid Image reconstruction Challenge

all of this should go

—
Reply to this email directly, view it on GitHub, or unsubscribe.

You are receiving this because you are subscribed to this thread.

Casper da Costa-Luis

unread,

Jan 23, 2026, 2:22:15 PMJan 23

to SyneRBI/SIRF-Contribs, Subscribed

casperdcl left a comment (SyneRBI/SIRF-Contribs#28)

It was 281 commits before I filtered out all unnecessary files.

—
Reply to this email directly, view it on GitHub, or unsubscribe.

You are receiving this because you are subscribed to this thread.

Casper da Costa-Luis

unread,

Jan 24, 2026, 4:04:41 PMJan 24

to SyneRBI/SIRF-Contribs, Push

@casperdcl pushed 55 commits.

ceec39b add SGD example
d9960e7 update deps
8f79c64 Merge pull request #23 from SyneRBI/sgd-example
88a91ad image quality callback
0151706 improve local run docs
40951c3 Merge pull request #25 from SyneRBI/metrics-and-docs
9bc2ed8 document --gpus all
b173181 added non-negativity
3706378 final cleanup
5257352 Merge pull request #35 from SyneRBI/ISTA
28ed970 slight doc updates
0c5c5a5 drop symlink
98862a7 Merge pull request #56 from SyneRBI/drop-symlink
425a619 raise runtime to 10 mins
3e5f107 Merge pull request #75 from SyneRBI/bump-runtime
83fc94a add examples for common pip/conda packages
5e0501f add note on conda vs pip
6c1964a Merge pull request #76 from SyneRBI/example-conda-pip
e705493 merging main...
76a9d31 docker dev instructions
507a23a Merge pull request #81 from SyneRBI/docker-user-run
c963ea1 Merge branch 'main' into Siemens_mMR_ACR
2c345b3 docs: leaderboard config & metrics runtime
f23a6ec Merge pull request #83 from SyneRBI/doc-leaderboard-metrics
b7ebd0d docs: slight tidy
fb32add Merge pull request #79 from SyneRBI/Siemens_mMR_ACR
867f5df mention wiki FAQs
2ac4d2d Merge branch 'main' into kris_eval
a79494c Merge remote-tracking branch 'origin/main' into kris_eval
26dcf5a Merge branch 'kris_eval' into Siemens_mMR_ACR2
f146e6c Merge pull request #89 from KrisThielemans/kris_eval
d396a74 Merge remote-tracking branch 'origin/main' into Siemens_mMR_ACR2
daf8c6b Merge pull request #88 from SyneRBI/Siemens_mMR_ACR2
b775506 Update README.md
d6e3ca2 Create LICENSE
18eb20d install pytorch-cuda & tensorflow-gpu
d0919d3 Merge pull request #112 from SyneRBI/gpu-libs
519496f resolve merge conflicts
0d2afaa update README
3b73b23 add ALG1 files
d036e98 update README
28409a8 use provided FOV mask
ff16048 remove MaxIteration callback
131f2d6 code cleanup + documentation
5397862 correct alg1 description in README
bc784c1 add ALG2 desctription to README
4453f79 Update README.md
5707d02 merge of paper branch that contains new simulations (#3)
2a64562 Update README.md (#4)
af12e82 fix use of to_device (#6)
4139176 Cupy fallback (#9)
aba33e3 update RDP (boundary conditions) such that they implementation is the exact same (up to float precision) with the one used in STIR's cuda relative difference prior (#11)
3499cf7 few memory optimisations
fe40b45 tidy for merge into SIRF-Contribs
0ae6241 add MaGeZ

—
View it on GitHub or unsubscribe.
You are receiving this because you are subscribed to this thread.

Kris Thielemans

unread,

Jan 25, 2026, 7:10:58 AMJan 25

to SyneRBI/SIRF-Contribs, Subscribed

@KrisThielemans requested changes on this pull request.

Many thanks @casperdcl this (and the history) is looking a lot better.

@gschramm could you (suggest) update(s for) the README and LICENSE?

In src/Python/sirf/contrib/MaGeZ/README.md:

> @@ -0,0 +1,52 @@
+# MaGeZ contribution to the PETRIC reconstruction challenge
+
+This repository contains algorithms developed by the MaGeZ team submitted to
+the [2024 PETRIC reconstruction challenge](https://github.com/SyneRBI/PETRIC) and
+published in the preprint [Fast PET Reconstruction with Variance Reduction and Prior-Aware Preconditioning](https://arxiv.org/abs/2506.04976).

@gschramm I suppose it's best to refer to https://github.com/SyneRBI/PETRIC-MaGeZ here and to the published paper as opposed to the preprint?

In src/Python/sirf/contrib/MaGeZ/README.md:

> +- Matthias Ehrhardt, Univeristy of Bath, United Kingdom
+- Georg Schramm, KU Leuven, Belgium
+- Zeljko Kereta, Univeristy College London, United Kingdom

"Univeristy" spelling

In src/Python/sirf/contrib/MaGeZ/README.md:

> +## Simulation results related to 2024 PETRIC and our paper
+
+To reproduce all simulation results checkout the tag `2024_paper_simulation_results`
+and have a look at the [Readme.md in the simulation_src folder](https://github.com/SyneRBI/PETRIC-MaGeZ/blob/main/simulation_src/README.md).

this should be deleted (or rephrased in terms of https://github.com/SyneRBI/PETRIC-MaGeZ, but I don't think that's appropriate here).

In src/Python/sirf/contrib/MaGeZ/README.md:

> +**To reproduce the PETRIC results of our three submitted algorithms
+(ALG1, ALG2, ALG3), please checkout the respective tag.**

@gschramm I thnk this should be deleted, as well as text below on ALG2, ALG3, unless they get added

—
Reply to this email directly, view it on GitHub, or unsubscribe.

You are receiving this because you are subscribed to this thread.

Edoardo Pasca

unread,

Jan 26, 2026, 4:12:47 AMJan 26

to SyneRBI/SIRF-Contribs, Subscribed

@paskino commented on this pull request.

In src/Python/sirf/contrib/MaGeZ/main.py:

> +        self._fov_mask = data.FOV_mask
+
+        # add a small number in the adjoint ones outside the FOV to avoid NaN in division
+        self._adjoint_ones.maximum(1e-6, out=self._adjoint_ones)
+
+        # initialize list / ImageData for all subset gradients and sum of gradients
+        self._summed_subset_gradients = self.x.get_uniform_copy(0)
+        self._subset_gradients = []
+
+        self._complete_gradient_epochs = complete_gradient_epochs
+
+        self._precond_update_epochs = precond_update_epochs
+
+        # setup python re-implementation of the RDP
+        # only used to get the diagonal of the RDP Hessian for preconditioning!
+        # (diag of RDP Hessian is not available in SIRF yet)

@KrisThielemans @evgueni-ovtchinnikov do you know whether this is still true?

The code could be significantly simplified otherwise.

—
Reply to this email directly, view it on GitHub, or unsubscribe.

You are receiving this because you are subscribed to this thread.

Evgueni Ovtchinnikov

unread,

Jan 26, 2026, 6:41:34 AMJan 26

to SyneRBI/SIRF-Contribs, Subscribed

@evgueni-ovtchinnikov commented on this pull request.

In src/Python/sirf/contrib/MaGeZ/main.py:

> +        self._fov_mask = data.FOV_mask
+
+        # add a small number in the adjoint ones outside the FOV to avoid NaN in division
+        self._adjoint_ones.maximum(1e-6, out=self._adjoint_ones)
+
+        # initialize list / ImageData for all subset gradients and sum of gradients
+        self._summed_subset_gradients = self.x.get_uniform_copy(0)
+        self._subset_gradients = []
+
+        self._complete_gradient_epochs = complete_gradient_epochs
+
+        self._precond_update_epochs = precond_update_epochs
+
+        # setup python re-implementation of the RDP
+        # only used to get the diagonal of the RDP Hessian for preconditioning!
+        # (diag of RDP Hessian is not available in SIRF yet)

yes it is not available in SIRF yet

—
Reply to this email directly, view it on GitHub, or unsubscribe.

You are receiving this because you are subscribed to this thread.

Kris Thielemans

unread,

Jan 26, 2026, 8:34:31 AMJan 26

to SyneRBI/SIRF-Contribs, Subscribed

@KrisThielemans commented on this pull request.

In src/Python/sirf/contrib/MaGeZ/main.py:

> +        self._fov_mask = data.FOV_mask
+
+        # add a small number in the adjoint ones outside the FOV to avoid NaN in division
+        self._adjoint_ones.maximum(1e-6, out=self._adjoint_ones)
+
+        # initialize list / ImageData for all subset gradients and sum of gradients
+        self._summed_subset_gradients = self.x.get_uniform_copy(0)
+        self._subset_gradients = []
+
+        self._complete_gradient_epochs = complete_gradient_epochs
+
+        self._precond_update_epochs = precond_update_epochs
+
+        # setup python re-implementation of the RDP
+        # only used to get the diagonal of the RDP Hessian for preconditioning!
+        # (diag of RDP Hessian is not available in SIRF yet)

It is in STIR, which is how we work around it in PETRIC2. https://github.com/SyneRBI/PETRIC2/blob/dc1370277e747ce6c29cee4484c19230819026c9/SIRF_data_preparation/run_LBFGSBPC.py#L101 and all the lines above to get around this problem. @evgueni-ovtchinnikov is there a SIRF issue already to expose this?

—
Reply to this email directly, view it on GitHub, or unsubscribe.

You are receiving this because you are subscribed to this thread.

Casper da Costa-Luis

unread,

Jan 27, 2026, 6:35:47 AMJan 27

to SyneRBI/SIRF-Contribs, Push

@casperdcl pushed 2 commits.

fa7372c make pip-installable
b7fe1c2 mention MaGeZ

—
View it on GitHub or unsubscribe.
You are receiving this because you are subscribed to this thread.

Edoardo Pasca

unread,

Jan 29, 2026, 5:52:51 AMJan 29

to SyneRBI/SIRF-Contribs, Subscribed

@paskino commented on this pull request.

In src/Python/sirf/contrib/MaGeZ/README.md:

> @@ -0,0 +1,52 @@
+# MaGeZ contribution to the PETRIC reconstruction challenge
+
+This repository contains algorithms developed by the MaGeZ team submitted to
+the [2024 PETRIC reconstruction challenge](https://github.com/SyneRBI/PETRIC) and
+published in the preprint [Fast PET Reconstruction with Variance Reduction and Prior-Aware Preconditioning](https://arxiv.org/abs/2506.04976).

⬇️ Suggested change

-published in the preprint [Fast PET Reconstruction with Variance Reduction and Prior-Aware Preconditioning](https://arxiv.org/abs/2506.04976).
+published in [Fast PET Reconstruction with Variance Reduction and Prior-Aware Preconditioning](https://doi.org/10.3389/fnume.2025.1641215).

—
Reply to this email directly, view it on GitHub, or unsubscribe.

You are receiving this because you are subscribed to this thread.

Edoardo Pasca

unread,

Jan 29, 2026, 5:53:12 AMJan 29

to SyneRBI/SIRF-Contribs, Subscribed

@paskino commented on this pull request.

In src/Python/sirf/contrib/MaGeZ/README.md:

> @@ -0,0 +1,52 @@
+# MaGeZ contribution to the PETRIC reconstruction challenge
+
+This repository contains algorithms developed by the MaGeZ team submitted to
+the [2024 PETRIC reconstruction challenge](https://github.com/SyneRBI/PETRIC) and
+published in the preprint [Fast PET Reconstruction with Variance Reduction and Prior-Aware Preconditioning](https://arxiv.org/abs/2506.04976

).
+
+## Authors
+
+- Matthias Ehrhardt, Univeristy of Bath, United Kingdom
+- Georg Schramm, KU Leuven, Belgium
+- Zeljko Kereta, Univeristy College London, United Kingdom

⬇️ Suggested change

-- Zeljko Kereta, Univeristy College London, United Kingdom
+- Zeljko Kereta, University College London, United Kingdom

—
Reply to this email directly, view it on GitHub, or unsubscribe.

You are receiving this because you are subscribed to this thread.

Edoardo Pasca

unread,

Jan 29, 2026, 6:00:07 AMJan 29

to SyneRBI/SIRF-Contribs, Subscribed

@paskino commented on this pull request.

In src/Python/sirf/contrib/MaGeZ/README.md:

> +This repository contains algorithms developed by the MaGeZ team submitted to
+the [2024 PETRIC reconstruction challenge](https://github.com/SyneRBI/PETRIC) and

⬇️ Suggested change

-This repository contains algorithms developed by the MaGeZ team submitted to
-the [2024 PETRIC reconstruction challenge](https://github.com/SyneRBI/PETRIC) and
+This repository contains the algorithm [ALG1](https://github.com/SyneRBI/PETRIC-MaGeZ/blob/ALG1/main.py) developed by the MaGeZ team submitted to
+the [2024 PETRIC reconstruction challenge](https://github.com/SyneRBI/PETRIC) and

—
Reply to this email directly, view it on GitHub, or unsubscribe.

You are receiving this because you are subscribed to this thread.

Edoardo Pasca

unread,

Jan 29, 2026, 6:00:59 AMJan 29

to SyneRBI/SIRF-Contribs, Subscribed

@paskino commented on this pull request.

In src/Python/sirf/contrib/MaGeZ/README.md:

> +
+## Simulation results related to 2024 PETRIC and our paper
+
+To reproduce all simulation results checkout the tag `2024_paper_simulation_results`
+and have a look at the [Readme.md in the simulation_src folder](https://github.com/SyneRBI/PETRIC-MaGeZ/blob/main/simulation_src/README.md).
+
+## Algorithms sumbitted to the 2024 PETRIC challenge
+
+**To reproduce the PETRIC results of our three submitted algorithms
+(ALG1, ALG2, ALG3), please checkout the respective tag.**

⬇️ Suggested change

-
-## Simulation results related to 2024 PETRIC and our paper
-
-To reproduce all simulation results checkout the tag `2024_paper_simulation_results`
-and have a look at the [Readme.md in the simulation_src folder](https://github.com/SyneRBI/PETRIC-MaGeZ/blob/main/simulation_src/README.md).
-
-## Algorithms sumbitted to the 2024 PETRIC challenge
-
-**To reproduce the PETRIC results of our three submitted algorithms
-(ALG1, ALG2, ALG3), please checkout the respective tag.**

—
Reply to this email directly, view it on GitHub, or unsubscribe.

You are receiving this because you are subscribed to this thread.

Edoardo Pasca

unread,

Jan 29, 2026, 6:04:20 AMJan 29

to SyneRBI/SIRF-Contribs, Subscribed

@paskino commented on this pull request.

In src/Python/sirf/contrib/MaGeZ/README.md:

> +### ALG2 (ALG2 branch, ALG2 tag)
+
+Similar to ALG1 but:
+The stepsize is set to 2.2 in the first epoch, and is then computed using the Barzilai-Borwein rule as described in Algorithm 2 in [https://arxiv.org/abs/1605.04131](https://arxiv.org/abs/1605.04131),
+with m = 1 but using the short stepsize (see [https://en.wikipedia.org/wiki/Barzilai-Borwein_method]([https://en.wikipedia.org/wiki/Barzilai-Borwein_method])) adapted to preconditioned gradient ascent methods
+
+### ALG3 (ALG3 branch, ALG3 tag)
+
+Using the same setup as in ALG2 but with two minor differences.
+First, we use a slightly smaller number of subsets.
+Second, we use a non-stochastic selection of subset indices by considering the cyclic group corresponding to the given number of subsets, finding all of its generators (i.e. the set of all coprimes of the number of subsets that are smaller), and then creating indices by consider specific generators at a time.

⬇️ Suggested change

-### ALG2 (ALG2 branch, ALG2 tag)
-
-Similar to ALG1 but:
-The stepsize is set to 2.2 in the first epoch, and is then computed using the Barzilai-Borwein rule as described in Algorithm 2 in [https://arxiv.org/abs/1605.04131](https://arxiv.org/abs/1605.04131),
-with m = 1 but using the short stepsize (see [https://en.wikipedia.org/wiki/Barzilai-Borwein_method]([https://en.wikipedia.org/wiki/Barzilai-Borwein_method])) adapted to preconditioned gradient ascent methods
-
-### ALG3 (ALG3 branch, ALG3 tag)
-
-Using the same setup as in ALG2 but with two minor differences.
-First, we use a slightly smaller number of subsets.
-Second, we use a non-stochastic selection of subset indices by considering the cyclic group corresponding to the given number of subsets, finding all of its generators (i.e. the set of all coprimes of the number of subsets that are smaller), and then creating indices by consider specific generators at a time.
+### Further info on MaGeZ submission
+
+Please refer to the original repository by the MaGeZ team for further information and notes on the submitted algorithms: [ALG1](https://github.com/SyneRBI/PETRIC-MaGeZ/tree/ALG1), [ALG2](https://github.com/SyneRBI/PETRIC-MaGeZ/tree/ALG2), [ALG3](https://github.com/SyneRBI/PETRIC-MaGeZ/tree/ALG3) .

—
Reply to this email directly, view it on GitHub, or unsubscribe.

You are receiving this because you are subscribed to this thread.

Edoardo Pasca

unread,

Jan 29, 2026, 6:07:14 AMJan 29

to SyneRBI/SIRF-Contribs, Subscribed

@paskino requested changes on this pull request.

A few changes for the README discussed with @KrisThielemans

—
Reply to this email directly, view it on GitHub, or unsubscribe.

You are receiving this because you are subscribed to this thread.

Kris Thielemans

unread,

Jan 29, 2026, 7:03:05 AMJan 29

to SyneRBI/SIRF-Contribs, Subscribed

@KrisThielemans commented on this pull request.

In src/Python/sirf/contrib/MaGeZ/LICENSE:

> +      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]

@gschramm, last chance before we merge :-)

—
Reply to this email directly, view it on GitHub, or unsubscribe.

You are receiving this because you are subscribed to this thread.

Kris Thielemans

unread,

Jan 29, 2026, 7:07:48 AMJan 29

to SyneRBI/SIRF-Contribs, Subscribed

@KrisThielemans commented on this pull request.

In src/Python/sirf/contrib/MaGeZ/README.md:

> @@ -0,0 +1,52 @@
+# MaGeZ contribution to the PETRIC reconstruction challenge
+
+This repository contains algorithms developed by the MaGeZ team submitted to
+the [2024 PETRIC reconstruction challenge](https://github.com/SyneRBI/PETRIC) and
+published in the preprint [Fast PET Reconstruction with Variance Reduction and Prior-Aware Preconditioning](https://arxiv.org/abs/2506.04976).
+
+## Authors
+
+- Matthias Ehrhardt, Univeristy of Bath, United Kingdom

⬇️ Suggested change

-- Matthias Ehrhardt, Univeristy of Bath, United Kingdom
+- Matthias Ehrhardt, University of Bath, United Kingdom

—
Reply to this email directly, view it on GitHub, or unsubscribe.

You are receiving this because you are subscribed to this thread.

Kris Thielemans

unread,

Jan 29, 2026, 7:09:53 AMJan 29

to SyneRBI/SIRF-Contribs, Push

@KrisThielemans pushed 1 commit.

384d5d9 Update README to make it appropriate for SIRF-Contribs

—
View it on GitHub or unsubscribe.
You are receiving this because you are subscribed to this thread.

Kris Thielemans

unread,

Jan 29, 2026, 7:11:27 AMJan 29

to SyneRBI/SIRF-Contribs, Subscribed

@KrisThielemans commented on this pull request.

In src/Python/sirf/contrib/MaGeZ/README.md:

> @@ -0,0 +1,35 @@
+# MaGeZ contribution to the PETRIC reconstruction challenge
+
+This repository contains the algorithm [ALG1](https://github.com/SyneRBI/PETRIC-MaGeZ/blob/ALG1/main.py) developed by the MaGeZ team submitted to
+the [2024 PETRIC reconstruction challenge](https://github.com/SyneRBI/PETRIC) (with minor edits by @paskino to reduce the use of video RAM) and
+published in [Fast PET Reconstruction with Variance Reduction and Prior-Aware Preconditioning](https://doi.org/10.3389/fnume.2025.1641215).
+
+## Authors
+
+- Matthias Ehrhardt, Univeristy of Bath, United Kingdom
+- Georg Schramm, KU Leuven, Belgium
+- Zeljko Kereta, University College London, United Kingdom
+
+
+### ALG1 (main branch, ALG1 tag)

⬇️ Suggested change

-### ALG1 (main branch, ALG1 tag)
+### ALG1 description

—
Reply to this email directly, view it on GitHub, or unsubscribe.

You are receiving this because you are subscribed to this thread.

Kris Thielemans

unread,

Jan 29, 2026, 7:11:57 AMJan 29

to SyneRBI/SIRF-Contribs, Push

@KrisThielemans pushed 1 commit.

4b2a19b final README fixes

—
View it on GitHub or unsubscribe.
You are receiving this because you are subscribed to this thread.

Kris Thielemans

unread,

Jan 29, 2026, 7:13:22 AMJan 29

to SyneRBI/SIRF-Contribs, Subscribed

@KrisThielemans approved this pull request.

feel free to squash the last 2 commits (or even squash that with another appropriate commit)

—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you are subscribed to this thread.

Casper da Costa-Luis

unread,

Jan 29, 2026, 9:40:36 AMJan 29

to SyneRBI/SIRF-Contribs, Subscribed

Merged #28 into master.

—
Reply to this email directly, view it on GitHub, or unsubscribe.

You are receiving this because you are subscribed to this thread.

Evgueni Ovtchinnikov

unread,

Feb 3, 2026, 1:06:10 PMFeb 3

to SyneRBI/SIRF-Contribs, Subscribed

@evgueni-ovtchinnikov commented on this pull request.

In src/Python/sirf/contrib/MaGeZ/main.py:

> +        self._fov_mask = data.FOV_mask
+
+        # add a small number in the adjoint ones outside the FOV to avoid NaN in division
+        self._adjoint_ones.maximum(1e-6, out=self._adjoint_ones)
+
+        # initialize list / ImageData for all subset gradients and sum of gradients
+        self._summed_subset_gradients = self.x.get_uniform_copy(0)
+        self._subset_gradients = []
+
+        self._complete_gradient_epochs = complete_gradient_epochs
+
+        self._precond_update_epochs = precond_update_epochs
+
+        # setup python re-implementation of the RDP
+        # only used to get the diagonal of the RDP Hessian for preconditioning!
+        # (diag of RDP Hessian is not available in SIRF yet)

is there a SIRF issue already to expose this?

Now there is: issue #1369, fixed by PR #1370, tested by pSTIR/tests/tests_qp_lc_rdp.py.