Holotomo improvements
-
enable half-tomo reconstruction (with the rotation center on the side). Need more details about the example dataset -
enable reconstructing a larger field of view - for ID16A: using the last distance field of view - would require masking the data ? Or are longer distances used to pad shorter ones ? - See
use_real_padding
andpadding_mix_factor
(20 pixel transition) in holoct - would need to reorganise the
load_align_zoom
function:- Compute image size based on chosen plane for the final field of view
- Load all projections and distances
- Zoom but don't pad
- Align all zoomed projections
- Pad:
- using real data (farther propagation planes)
- taking into account the correct alignment
- pad in the same way the direct beam images
- need to normalise the intensity, at least between planes for a given projection
- See
-
distorsion correction: -
finish the distortion calculation on GPU -
use a distortion calculation on CPU: almost as fast for the registration, and what takes time is the actual distortion using 3rd-order splines.. -
Optionally: use multiple probe modes (coherent) or a distorted probe ?
-
-
diagnostic/check of displacement: -
need to be able to display the 4 distances for multiple projections, and switch between distances quickly (no delay, no white image) -
either save normalised & shifted images (e.g. every 100 frames), or have a small GUI to display them ? -
At least some indicator should be written to the log file if something goes wrong - e.g. measure the std deviation between the raw and fitted values -
This could probably be done using a display of different images (maybe with a high-pass filter ?) using transparency (e.g. using silx), or by displaying the difference between images ?
-
-
do automatically AP**2 (probe only) as part of CTF / Paganin (c76400f5) -
VSNR removal - 3 or 4 iterations are OK -
Add growing-window option to the CoR search -
Add double-flatfield distortion correction (04f4d1eb) -
add optional filters for the double-flatfield: on the normalised projections (iobs/ref), perform a high-pass (>0.95) in the projection 2D plane, then low-pass along the projections index (or just the average). Then subtract (from the observed array) the resulting array multiplied by the reference array
-
-
Add Munch deringer correction -
Add export of phased projections to the NXTomo format (see e.g. what is the result of converting the phases from holoct using nxtomomill edf2nx data/prefix_ prefix.nx
whereprefix_
is the folder name with the edf phased images)- option now exists but actual reconstruction with nabu/tomwer needs to be tested
-
NXTomo: when exporting to the NXtomo format, either write in the log the command-line to use to re-do the reconstruction using the command-line version of nabu, or write a corresponding config file -
add command-line parameters for the CTF and Paganin alpha filter -
allow to reconstruct a subset of the available projections, using a similar option to ptycho's moduloframe
, to be able to use even/odd frames only etc (e.g. for denoising) -
investigate issues for samples with a high refraction index contrast ; see constrain_grad and attenuate options in holoct -
CoR: add option to refine the CoR based on the contrast on a small range of +/- n pixels -
CoR: add option to export a few slices reconstructed with different CoR over a small range of pixel shifts -
Add export to jpeg2k format (save space)
Edited by Vincent Favre-Nicolin