Added jupyter notebook to make parameter specification and running of the fusion easier #4
Conversation
…re specified in a separate jupyter notebook, where they are exported as config file (.json). `mvregfus_run.py` now only contains functions that _run_ the fusion using this config file. Added `run_fusion.ipynb` which contains a description of the fusion parameters, together with some functionality to generate a config file and to run the fusion. Added some convenience functionality: * Parameter `filepaths` now also takes a string. If so, glob.glob(string + '/*.czi') is used to find files in the specified folder (string). * Added parameter generate_file_order. If True, a `file_order.txt` file is generated in the `out_dir`. It contains the contents of the `filepaths` variable. This is useful, since MVRegfus currently renames the output files to mv_000_.... from which it is not possible to reconstruct which file was processed. * Parameter `registration_pairs` can be set to None. * If `registration_pairs == None`, the new parameters `n_volumes` and `n_views` are used to generate the registration pairs.
|
Ciao Marvin, |
max-brambach
changed the title
mvregfus_run.py such that now fusion parameters and paths a…
|
That's super cool! Thanks a lot for sharing. You're completely right that it makes sense to separate execution and configuration. And it's been on my todo list forever 😂. Exactly as you propose, I think it makes sense to define a parameter dictionary which can then be saved as a Regarding the implementation, probably the |
| if generate_file_order: | ||
| with open(out_dir + '/file_order.txt', 'w') as f: | ||
| for item in filepaths: | ||
| f.write("%s\n" % item) |
There was a problem hiding this comment.
Definitely the output filenames should reference the input files.. The easiest way will be to simply change this in mv_graph.py, which I'll do.
| # ----- other parameters ---- | ||
| # options for DCT image quality metric for fusion | ||
| # setting None automatically calculates good values | ||
| # size of the cubic volume blocks on which to calc quality | ||
| dct_size = None | ||
| # size of maximum filter kernel | ||
| dct_max_kernel = None | ||
| # size of gaussian kernel | ||
| dct_gaussian_kernel = None | ||
|
|
||
| # weight normalisation parameters | ||
| # normalise such that approx. <dct_cumulative_weight_best_views> weight is | ||
| # contained in the <dct_how_many_best_views> best views | ||
| dct_how_many_best_views = 2 | ||
| dct_cumulative_weight_best_views = 0.9 | ||
|
|
||
| # options for weighted Lucy Richardson multi-view deconvolution | ||
| # maximum number of iterations | ||
| LR_niter = 25 # iters | ||
| # convergence criterion | ||
| LR_tol = 5e-5 # tol | ||
| # gaussian PSF sigmas | ||
| LR_sigma_z = 4 # sigma z | ||
| LR_sigma_xy = 0.5 # sigma xy | ||
|
|
||
|
|
||
| # how to calculate final fusion volume | ||
| # 'sample': takes best quality z plane of every view to define the volume | ||
| # 'union': takes the union of all view volumes | ||
| final_volume_mode = 'union' | ||
| # whether to perform an affine chromatic correction | ||
| # and which channel to use as reference | ||
| perform_chromatic_correction = False | ||
| ref_channel_chrom = 0 | ||
|
|
||
|
|
||
| # ----- derivatives of the set parameters ----- | ||
| channelss = [channels]*len(filepaths) | ||
| reg_channels = [reg_channel] *len(filepaths) | ||
| ref_views = [ref_view] *len(filepaths) | ||
| registration_pairss = [registration_pairs] *len(filepaths) | ||
| view_dict = {i:{'view':i, 'ill':2} for i in list(range(16))} | ||
| # where elastix can be found (top folder) ## urrently not used | ||
| elastix_dir = '/data/shared/elastix' |
There was a problem hiding this comment.
All these would probably need default values in the function definition (or default values in a json file for default values).
There was a problem hiding this comment.
Yes, I've only selected the ones I'm using normally, but of course it makes sense to have the other parameters available as well.
| n_volumes = parameters['n_volumes'] | ||
| n_views = parameters['n_views'] |
There was a problem hiding this comment.
These can probably be automatically determined in mv_graph.build_multiview_graph.
Changed
mvregfus_run.pysuch that now fusion parameters and paths are specified in a separate jupyter notebook, where they are exported as config file (.json).mvregfus_run.pynow only contains functions that run the fusion using this config file.Added
run_fusion.ipynbwhich contains a description of the fusion parameters, together with some functionality to generate a config file and to run the fusion.Added some convenience functionality:
filepathsnow also takes a string. If so, glob.glob(string + '/*.czi') is used to find files in the specified folder (string).file_order.txtfile is generated in theout_dir. It contains the contents of thefilepathsvariable. This is useful, since MVRegfus currently renames the output files to mv_000_.... from which it is not possible to reconstruct which file was processed.registration_pairscan be set to None.registration_pairs == None, the new parametersn_volumesandn_viewsare used to generate the registration pairs.