NGFF-EMBL-HD-2020-10-02

--- tags: NGFF --- # NGFF-EMBL-HD-2020-10-02 Attending: Martin Schorb, Nicholas Chiaruttini, Constantin Pape, Christian Tischer, Josh Moore - Tischi: - Definition of "Dataset" - NGFF: could we show some community spirit. Stop doing own thing - mobie definition of "Dataset" is a bunch of images that make sense to be displayed in one context. Correlated / HCS. - .. but tomograms from same physical specimen (random grid). Got lost. - Nicholas: general problem. What do you want to solve? Too general to work on. - FLIM dataset where each pixel is 3000 features. How would you show it? - How do you align a sparse dataset? Lots of little cubes. - Constantin: for tomograms, it would be on a regular grid - Josh: spec "subclassing" - Martin: in CLEM you have the pyramid, ... and point to it via a bookmark - Nicolas: have a raster image. combining it with vector data. Need **SVG** or something. - Martin: annotations are another layer. Nicolas: a mess. - Tischi: got a bit loss trying to find - Nicolas: chaining of reference (global --> ) - Josh: like a(n infinite) desktop would let you order collections visually - Side-note about ... (Josh: I forgot what this was) - Goal : - Josh: convince the microscope vendor to use ngff, last vendor meeting : no easy solution - IDR side-node: - Martin: 2 issues - (1) Binary: S3 COVID. Using N5, but doesn't matter. Was in touch with the EMPIAR guys (https://www.ebi.ac.uk/pdbe/emdb/empiar/). - Josh: cover Zarr/N5 and IDR/EMPIAR relationships - (2) ... ? - Tischi: off-topic? But how do we submit to EMPIAR? - Constantin: common-denominator is "What format should we use?" - Martin: BDV playground was concentrating on the physical coordinates as the driving information. Now have an additional layer of display layout. (e.g. 20 volumes on screen). What's the hierarchy of the layout and the physical info? Wouldn't consider them independent but layers... Viewer has a projection of the virtual physical space and projects it to one pixel on the viewer screen. This would add an additional layer between pixel on the screen and the physical location. - Nicolas: have something correct in Bio-Formats with pixel/physical size - missing affine transform in 3D. - then for another layer you have physical-- chaining affine transforms - but looking at **blender** you know how to position in 3D space with some hierarchy between objects. - Also: how to position in time? Lacking in Bio-Formats. Start time? End time? And if not consistent. - Usually isotropically in space, but less likely in time. - Tischi: advantage of vector is you don't have to decide where (0,0,0) is. relative distances. - Nico: in data viewer is done on its own without - would have been awesome (for 3D) to have grid & SVG. annotations/ROIs is potentially too difficult. - Tischi: summary- if in mobie we have JSON with one or more layout options like: - "use this affine transform" (list for a bunch of objects) - "random grid" - Martin: will need to specify some stuff for different sized objects. - ... "SVG" - Constantin: should be relative easily for the grid dataset - Tischi: don't think this has to be one Zarr. For the tomogram. - Josh: consistent way to load metadata & arrays - Tischi: most important is the loading of remote "links" to other containers - Josh: chicken-n-egg on the implementation - people don't want to implement until the spec is done. - but would be good to have a **few** people - Martin: want one file with a bookmark or redirector. Tischi: perfect - Nicolas: need static dataset or git... - Versioning: - Martin: mismatch should be a warning - Josh: good longer discussion! :smile: - Specs - Josh: good to discuss. Can you write up the mobie one? This time plus a doc for how to write up a spec. - Example from another community https://github.com/cf-convention/discuss/issues/62#issuecomment-701422957 - Nico: need a way to save chains of layouts, in registering brain mouse context. (Have **JSON**) - https://github.com/bigdataviewer/bigdataviewer-playground - Martin: like https://github.com/saalfeldlab/render? - Nico: sounds like it is compatible. saving transformations. - Christian: **"Club of the Willing"** :100: