Use phase data in main workflow and add option for phase noise map #10
Conversation
|
I'm going to put this in draft mode until I know what the scaling situation should be for the phase noise volumes. |
|
Ah I remember having the same issue. If no phase unwrapping has be done, it's hard. The header does not necessarily contains the information (even in the DICOM, I remember having to look into a siemens raw data .dat file), and it's probably vendor specific. A layman solution would be to let the user provide this to us. |
|
I didn't think NORDIC_Raw unwrapped the phase data.
Sorry, which information? The scale of the phase data? What about rescaling the phase data and noise data in one array, then splitting it up? Not within patch-denoise of course, since the data are expected to be in radians already. |
tests/test_spacetime_utils.py
Outdated
| print(f"Seed: {seed}") | ||
| rng = np.random.RandomState(seed) | ||
| medium_random_matrix = rng.randn(200, 200, 100) | ||
| print(f"Mean of raw: {np.nanmean(medium_random_matrix)}") | ||
| print(f"Max of raw: {np.nanmax(medium_random_matrix)}") | ||
| print(f"Min of raw: {np.nanmin(medium_random_matrix)}") | ||
|
|
||
| noise_map = estimate_noise(medium_random_matrix, block_dim) | ||
| print(f"Mean of noise map: {np.nanmean(noise_map)}") | ||
| real_std = np.nanstd(medium_random_matrix) | ||
| print(f"SD of raw: {real_std}") | ||
| err = np.nanmean(noise_map - real_std) | ||
| print(f"Err: {err}") | ||
| assert err <= 0.1 * real_std |
There was a problem hiding this comment.
I'm digging into the intermittent test failures with the noise maps. The noise map sometimes has a really massive value. I'm not sure why.
Seed: 0
Mean of raw: 0.0004815311392502157
Max of raw: 4.7965495404996
Min of raw: -5.031764123854618
SD of raw: 0.9994477899078287
Mean of noise map: 134637028848.5359
Max of noise map: 527766886400.0
Min of noise map: 5e-323
SD of noise map: 2550814615827.277
Err: 134637028847.53642
There was a problem hiding this comment.
If the thermal noise volumes have values from -5 to 5, I can't see any way for the noise map to have values from 0 to 527766886400. I think there's a bug in the noise map estimation.
@paquiteau WDYT?
There was a problem hiding this comment.
I can't reproduce the error locally. I've tried with seeds going from 0 to 60 and block sizes from 5 to 10, but never get these massive noise map values.
There was a problem hiding this comment.
I'm thinking it must be something to do with the float precision.
|
We just need to concatenate the raw phase and noise phase, then rescale, then split. |
|
If you're okay merging then please do. I think the test errors can be addressed in a separate PR. |
Closes #8.
Changes proposed:
--noise-map-phase, for the phase component of the noise map.