functional writing into BigDataViewer/BigStitcher HDF5

[nvladimus]

Hi, I implemented writing of multiple rows into a single H5 file that contains rows as views grouped by:

• shutter configuration (Left/Right) as 'illumination' attribute
• laser as 'channel'
• rotation as 'angle' Tested in Demo mode, seems to work fine, BigStitcher reads the generated file correctly.

Remaining issue: when a row contains new rotation, the image writer does not finalize XML file. Maybe because rotation not implemented in Demo mode? Not sure.

[ffvoigt]

Awesome - I can get it to work, but I ran into an additional issue:

• Simple tiled acquisitions generated with the tiling wizard fail to work with an File "h5py/h5g.pyx", line 161, in h5py.h5g.create ValueError: Unable to create group (name already exists) error message - it seems to me that currently, no tile count is present. A while ago, I've implemented a simple tile count as the generate_tiledict method here.

Minor things:

• When running acquisition lists with rotations, the software prints a message that the microscope cannot move to the save rotation position, but sample rotation in an acquisition list works per se - the updated rotation is displayed in the coordinate readout. So, I don't know why the XML does not contain the rotation angle...
• mesoSPIM_ImageWriter.py has a lot of diagnostic print statements - I'm trying to slowly get away from them (mea culpa, see above)- btw: you can use logger.info('Put your info here') instead.
• in this commit, the local tmp directory seems to be missing - I wanted to keep it (and the small .txt in it) as a good default save location for demodata that works out of the box.

Questions for you:

• How shall we represent this in the GUI? Currently, it is not obvious that all rows need to have the same ~.h5, one option is to turn this into another option in the Filename Wizard?
• How do you want to handle shutter = Both? - People should be discouraged from using it, but there might be the need for a warning.

[nvladimus]

Thanks for your feedback, @ffvoigt. Here are some major changes ✌️ tiling is added and seems to work. ✨ tiles contain positional information (XYZ) - which places them on a 3D grid in BigStitcher. Each tile also knows its voxel size.

Minor things:

• rotation-related bug fixed, it was in npy2bdv library.
• my debugging print messages are removed, they were intended temporarily.
• I added /tmp directory to git .ignore list, using it for my own debugging purposes, sorry for confusion.

To you questions,

• Regarding the file names, I suggest to add a drop-down menu in acquisition manager that specifies extension of ALL files in the list: .raw or .h5
• The case shutter = Both is handled as illumination setting No.3, this is not a problem for BigStitcher. Whether user wants/needs it, is another question. Since it's already there, I see no reasons to remove it. Maybe useful e.g for some tricky alignment situations.

[ffvoigt]

So, changes are merged - Thanks -but there is still some weirdness going on: When dealing with a demo dataset with 1.6 micron pixel size, Bigdataviewer displays it as 1024 pixels across and not 1638 µm. As a result, a tiled dataset suddenly has gaps even though the overlap is correct. This might be an issue with the order of transformations in the XML or with the units of affine matrix transformation in the XML.

Also, I've added single h5 support in the filename wizard, I don't know whether having another option for all the files is a good idea, but I will think about it.

[nvladimus]

Good catch - I forgot that transformation coordinates are in px, not um. Fixed, will submit new PR now.