Creating more powerful support for opening tables

[tischi]

@constantinpape @jhennies

It is kind of hard to create a MoBIE project and also hard to (manually) modify the corresponding dataset.json.

I am thus pondering the idea to see how much of a dataset.json we could express in dataset.tsv, which would be much easier to create and modify. This will only be an alternative and only be able to represent a subset of all functionality, but I feel it could be very useful.

Some ideas for columns in this table:

• view_name: serves as the default view ID for that image or segmentation
• source_path:
• source_type: image, segmentation
• affine: 12 comma separated numbers
• grid_id: everything with the same grid_id would become a grid view
• ui_group:
• contrast_limits: min, max
• invert:
• color:
• view_id: everything with the same view_id would be joined into one view

Writing this up I feel one could express quite a bunch of stuff....

Any thoughts?

[tischi]

I would need some input how to best handle multi-channel datasets.
There seem to be two options:

1. One row per multi-channel dataset

• This would be easier as one needs to add less information to the table; for instance typically one would transform all channels with the same transformation; however typically one would need different colours and contrast limits, which would be annoying

1. One row per single channel

• This would be closer to what we currently have in the dataset.json; it would either require adding another column channel_index or we could establish some convention where the URI could have an ;ch=0 suffix. Advantages would be that one could readily specify display setting and would could also have different transforms for different channels, which can be useful to align some shifts as they can occur in multi-color imaging

What do you think @constantinpape ?

[tischi]

Here's a prototype specification.