search

python-microscopy / python-microscopy

The python-microscopy environment - an open-source package for light microscopy
http://www.python-microscopy.org
GNU General Public License v3.0
83 stars 25 forks source link

[ENHANCEMENT] concatenate tiled series #581

Closed barentine closed 3 years ago

barentine commented 3 years ago

Is your feature request related to a problem? Please describe. I have a lot of tiled overview images that I want to view as one supertile source (just for viewing at the moment)

Potential soutions

pyme-cluster:// - aggregate ImageOutput module: Add to HTTPDataServer aggregate_h5 endpoint such that we could enable pyme-cluster:// - aggregate scheme for output.ImageOutput. This would not entail much in the dataserver, but rather in IO.h5File.H5File, where put_file would need to have a bit more logic on pzf frames (since frame numbers will overlap).

However, because PointScanner events, ScannerXPos, ScannerYPos only include time stamps and not frame numbers, and we use mdh['StartTime'] to map to frames, we would need to spoof the event times when appending them and associated frames (would also need a lock taken out in part of H5File.put_file to update a max_frame / max_time variable in case we try and push frames from multiple sources at once. This might be easiest in a separate method to put_file that can take events, frames, and metadata all at the same time.

SuperClusterPZFDataSource: A datasource for the superlative extraordinaire who wants to make a SupertileDataSource from only the finest input datasource. Would be grosser from a ui perspective in my opinion, as the other I can make happen behind the scenes easily in a recipe I'm already running for detection/queueing on individual tile overviews, while this would I would somehow need to generate an arbitrarily long list of inputs for a recipe module.

tile_super_pyramid: A method in tile_pyramid which directly takes as input a list of tile series and generates a mondo pyramid, never writing out a singular full concatenated tile series.

modify the tile viewer to accept multiple inputs might be the most pragmatic if not least powerful

Are you prepared to work on it? always

barentine commented 3 years ago

modifying the tileviewer might be super simple from the openseadragon perspective as we just need to add more tile sources.

adding a multitileviewer.py and associated multitileviewer.html really might be straightforward

David-Baddeley commented 3 years ago

I think any aggregation of separate tile pyramids is going to be a bit tricky due to the fact that it's highly unlikely that the tile origins will align on the same grid. The 3 options I'd consider would be:

a) save everything to the one pyramid (this could be done from separate, per-well protocols, but ultimately all writing into the same pyramid structure). This would ensure tiles are aligned and make viewing a cinch. Might not be practically realisable until we have a distributed version of the tile pyramid though.

b) do something on the viewer end to load and display separate tile pyramids in the one view. May not be super simple due to the potential for tile boundary misalignment.

c) if just for an overview for publication / presentations, pull the top level tiles from each pyramid and write them into an image in a custom script / ipython notebook

David-Baddeley commented 3 years ago

d) display N separate pyramids, one for each well (or display a non-zoomable overview as in c, and have clicking on a specific well open that well in the openseadragon viewer)

barentine commented 3 years ago

After discussion with David: Only need at the moment is for display in presentations / publication. Will write a jupyter notebook to load a list of pyramids as top-layer supertiles and generate the super supertile image we're looking for. Will PR the notebook to https://github.com/python-microscopy/example_notebooks

barentine commented 3 years ago

now taken care of in an example notebook