learning of parameter maps for tv reconstruction

[koflera]

we here try to include basic code for being able to run the method described in

https://arxiv.org/abs/2301.05888

using MRpro in an "advanced example" or so

[github-actions]

Coverage Report

File	Stmts	Miss	Cover	Missing
src/mrpro
_version.py	6	2	67%	7–8
src/mrpro/algorithms/csm
inati.py	25	1	96%	43
src/mrpro/algorithms/dcf
dcf_voronoi.py	55	4	93%	15, 55–56, 89
src/mrpro/algorithms/optimizers
adam.py	30	6	80%	108, 125–129
cg.py	56	1	98%	139
pdhg.py	81	3	96%	178–179, 185
pgd.py	53	4	92%	107, 152–155
src/mrpro/algorithms/reconstruction
DirectReconstruction.py	28	6	79%	62, 65, 70, 77–79
IterativeSENSEReconstruction.py	13	1	92%	79
Reconstruction.py	50	13	74%	54–56, 80–87, 109, 112
RegularizedIterativeSENSEReconstruction.py	51	26	49%	104–108, 122–161
src/mrpro/data
AcqInfo.py	171	7	96%	49, 56, 134–135, 137, 243, 367
CsmData.py	47	5	89%	17–19, 233–235
Dataclass.py	314	27	91%	59, 320, 336, 402, 460–462, 475, 570, 590–591, 593, 608–609, 611, 645, 658–659, 664–665, 832–833, 858, 865, 870–871, 873
DcfData.py	35	2	94%	15, 62
EncodingLimits.py	97	3	97%	37, 127, 130
IData.py	52	2	96%	119, 133
IHeader.py	127	7	94%	68–71, 250, 254, 258, 262
KData.py	228	24	89%	122–123, 138, 145, 156–167, 178, 186, 197, 236, 258–260, 304–305, 377, 543, 545, 616
KHeader.py	178	14	92%	31, 115–121, 148, 196, 203–204, 231–238
KNoise.py	22	1	95%	44
KTrajectory.py	95	3	97%	163, 165, 185
QData.py	32	1	97%	43
Rotation.py	764	42	95%	104, 202, 339, 437, 481, 499, 586, 588, 597, 631, 633, 696, 773, 778, 781, 796, 813, 818, 894, 1082, 1087, 1090, 1114, 1118, 1142, 1262, 1264, 1272–1273, 1337, 1419, 1623, 1630–1632, 1691, 1787, 1939, 1974, 1978, 2154, 2175
SpatialDimension.py	240	20	92%	34, 103, 146, 152, 158, 278–280, 293–295, 329, 347, 360, 373, 386, 399, 408–409, 437
src/mrpro/data/traj_calculators
KTrajectoryCalculator.py	26	2	92%	84, 95
KTrajectoryCartesian.py	28	4	86%	107–110, 114
KTrajectoryIsmrmrd.py	19	1	95%	57
KTrajectorySpiral2D.py	57	13	77%	63–66, 69, 71, 73, 75, 77, 105, 107, 134–136
src/mrpro/operators
AveragingOp.py	34	2	94%	53, 78
CartesianSamplingOp.py	108	4	96%	143, 182, 257, 361
ConjugateGradientOp.py	91	10	89%	35–43, 62, 64, 100, 106, 211, 213, 216
ConstraintsOp.py	79	4	95%	78, 80, 211, 216
EndomorphOperator.py	32	2	94%	52, 58
FiniteDifferenceOp.py	27	2	93%	40, 105
FourierOp.py	97	6	94%	169–170, 189, 234, 272, 277
Functional.py	71	4	94%	17–18, 116, 118
GridSamplingOp.py	163	15	91%	72–73, 82–83, 90–91, 94, 96, 98, 282, 290–291, 303, 309–310
LinearOperator.py	201	7	97%	107, 217, 255, 296, 305, 313, 330
LinearOperatorMatrix.py	172	19	89%	82, 119, 152, 161, 166, 175–178, 191–194, 202–203, 208–209, 221, 324, 354, 381
MultiIdentityOp.py	13	2	85%	43, 48
NonUniformFastFourierOp.py	190	10	95%	69, 96, 206, 208, 242, 244, 321, 379, 429, 434
Operator.py	89	4	96%	32, 89, 122, 132
OptimizerOp.py	65	3	95%	48–57
PatchOp.py	47	3	94%	81, 112, 127
ProximableFunctionalSeparableSum.py	64	3	95%	107, 198, 209
SliceProjectionOp.py	178	10	94%	45, 62, 64, 70, 153, 179, 215, 236, 269, 309
WaveletOp.py	119	5	96%	151, 169, 204, 209, 232
ZeroPadOp.py	16	1	94%	30
src/mrpro/operators/functionals
SSIM.py	71	7	90%	60–80, 82, 86, 114, 147
src/mrpro/operators/models
EPG.py	204	43	79%	85–100, 148–152, 172–175, 196–201, 264, 269, 285–287, 307–308, 313, 337, 342, 367, 372, 426, 483, 584, 611
src/mrpro/phantoms
brainweb.py	297	40	87%	276, 290–294, 349–359, 398, 454–457, 479–480, 485–486, 488–489, 493, 501, 508–509, 550, 621, 624–625, 644, 653–656, 667, 669, 700–701, 715, 723
fastmri.py	106	10	91%	50–51, 59, 65, 162, 169–171, 174–175, 189
m4raw.py	74	4	95%	58–59, 74, 76
mdcnn.py	71	7	90%	58, 62–63, 70, 82, 88, 135
src/mrpro/utils
RandomGenerator.py	156	15	90%	23–24, 36, 38, 188, 212, 428, 446, 528, 799, 829–832, 895, 898
filters.py	62	2	97%	44, 49
indexing.py	182	1	99%	321
pad_or_crop.py	32	6	81%	27, 31, 62, 65, 68, 71
reshape.py	139	9	94%	112, 306, 418–420, 441, 443, 450, 465
slice_profiles.py	49	6	88%	21, 37, 119–122, 155
split_idx.py	10	2	80%	43, 47
summarize.py	57	7	88%	40–41, 70–73, 77, 81
typing.py	73	56	23%	9–236
unit_conversion.py	72	16	78%	34, 44, 51, 53, 60, 62, 69, 71, 78, 80, 89, 100, 121, 123, 144, 146
TOTAL	7661	592	92%

Tests	Skipped	Failures	Errors	Time
2954	0 💤	0 ❌	0 🔥	2m 1s ⏱️

[github-actions]

📚 Documentation

📁 Download as zip
🔍 View online

[koflera]

except that the example is probably not nice yet etc, the main problem is that while training on the GPU, the GPU memory continues to increase, eventually leading to a CUDA out of memory error.
Is it super evident where the problem lies?

[koflera]

Finally, a first draft of the example is ready!
@fzimmermann89 I ended up undoing most of the changes that we had discussed. The most relevant were:

• normalization by std instead of the norm: normalizing by the std had the problem that when at the end of the example, we want to compare the iterative SENSE recon to the PDHG ones, the resulting scaling factor was substantially different when normalizing by the std. normalizing by the norm was much more consistent, so I went for that.
• initial value estimation as part of the network: given that approximately solving the normal equations only makes sense if we have multi-coil problem, I abstained from doing that and instead left that to the user to be done outside. I also think it makes the network more concise.
• creating the dataset with a torch.utils.data.TensorDataset and using the split function, required wo write an additional Wrapper to able to use the method prepare_data of the resulting SubSets. Additionally, after a few hours of trying to stop mypy from complaining, I gave up and went back to manually creating two different datasets.

[fzimmermann89]

sorry, you did not push yet, did you? :D

[koflera]

sorry, you did not push yet, did you? :D

no no, still waiting for the new training to finish... will probably push tomorrow in the morning ;)