Channel Equivalent For `Labels2DModel`

Nobody has claimed this yet.

Thanks for developing SpatialData! It's been a very ergonomic experience so far trying to convert a small subset of our lab's current multiplexed imaging pipeline as a MVP for testing.

One particular wishlist item would be an $N$-dimensional implementation of Labels2D, as for each SpatialImage a user may want to have several types of labels, such as:

Cell membrane segmentation masks
Nuclear membrane segmentation masks
Cluster Masks from various algorithms
etc...

Here is what I have currently:

Current `sdata` setup

SpatialData object with:
├── Images
│     ├── 'fov0': SpatialImage[cyx] (22, 512, 512)
│     ├── 'fov1': SpatialImage[cyx] (22, 1024, 1024)
│     ├── 'fov2': SpatialImage[cyx] (22, 1024, 1024)
│     ├── 'fov3': SpatialImage[cyx] (22, 1024, 1024)
│     ├── 'fov4': SpatialImage[cyx] (22, 1024, 1024)
│     ├── 'fov5': SpatialImage[cyx] (22, 1024, 1024)
│     ├── 'fov6': SpatialImage[cyx] (22, 1024, 1024)
│     ├── 'fov7': SpatialImage[cyx] (22, 1024, 1024)
│     ├── 'fov8': SpatialImage[cyx] (22, 1024, 1024)
│     ├── 'fov9': SpatialImage[cyx] (22, 1024, 1024)
│     └── 'fov10': SpatialImage[cyx] (22, 1024, 1024)
└── Labels
     ├── 'fov0_nuclear': SpatialImage[yx] (512, 512)
     ├── 'fov0_whole_cell': SpatialImage[yx] (512, 512)
     ├── 'fov1_nuclear': SpatialImage[yx] (1024, 1024)
     ├── 'fov1_whole_cell': SpatialImage[yx] (1024, 1024)
     ├── 'fov2_nuclear': SpatialImage[yx] (1024, 1024)
     ├── 'fov2_whole_cell': SpatialImage[yx] (1024, 1024)
     ├── 'fov3_nuclear': SpatialImage[yx] (1024, 1024)
     ├── 'fov3_whole_cell': SpatialImage[yx] (1024, 1024)
     ├── 'fov4_nuclear': SpatialImage[yx] (1024, 1024)
     ├── 'fov4_whole_cell': SpatialImage[yx] (1024, 1024)
     ├── 'fov5_nuclear': SpatialImage[yx] (1024, 1024)
     ├── 'fov5_whole_cell': SpatialImage[yx] (1024, 1024)
     ├── 'fov6_nuclear': SpatialImage[yx] (1024, 1024)
     ├── 'fov6_whole_cell': SpatialImage[yx] (1024, 1024)
     ├── 'fov7_nuclear': SpatialImage[yx] (1024, 1024)
     ├── 'fov7_whole_cell': SpatialImage[yx] (1024, 1024)
     ├── 'fov8_nuclear': SpatialImage[yx] (1024, 1024)
     ├── 'fov8_whole_cell': SpatialImage[yx] (1024, 1024)
     ├── 'fov9_nuclear': SpatialImage[yx] (1024, 1024)
     ├── 'fov9_whole_cell': SpatialImage[yx] (1024, 1024)
     ├── 'fov10_nuclear': SpatialImage[yx] (1024, 1024)
     └── 'fov10_whole_cell': SpatialImage[yx] (1024, 1024)

Ideally something like this would be great:

`sdata` with `C` equivalent

SpatialData object with:
├── Images
│     ├── 'fov0': SpatialImage[cyx] (22, 512, 512)
│     ├── 'fov1': SpatialImage[cyx] (22, 1024, 1024)
│     ├── 'fov2': SpatialImage[cyx] (22, 1024, 1024)
│     ├── 'fov3': SpatialImage[cyx] (22, 1024, 1024)
│     ├── 'fov4': SpatialImage[cyx] (22, 1024, 1024)
│     ├── 'fov5': SpatialImage[cyx] (22, 1024, 1024)
│     ├── 'fov6': SpatialImage[cyx] (22, 1024, 1024)
│     ├── 'fov7': SpatialImage[cyx] (22, 1024, 1024)
│     ├── 'fov8': SpatialImage[cyx] (22, 1024, 1024)
│     ├── 'fov9': SpatialImage[cyx] (22, 1024, 1024)
│     └── 'fov10': SpatialImage[cyx] (22, 1024, 1024)
└── Labels
     ├── 'fov0': SpatialImage[cyx] (2, 512, 512)
     ├── 'fov1': SpatialImage[cyx] (2, 1024, 1024)
     ├── 'fov2': SpatialImage[cyx] (2, 1024, 1024)
     ├── 'fov3': SpatialImage[cyx] (2, 1024, 1024)
     ├── 'fov4': SpatialImage[cyx] (2, 1024, 1024)
     ├── 'fov5': SpatialImage[cyx] (2, 1024, 1024)
     ├── 'fov6': SpatialImage[cyx] (2, 1024, 1024)
     ├── 'fov7': SpatialImage[cyx] (2, 1024, 1024)
     ├── 'fov8': SpatialImage[cyx] (2, 1024, 1024)
     ├── 'fov9': SpatialImage[cyx] (2, 1024, 1024)
     └── 'fov10': SpatialImage[cyx] (2, 1024, 1024)

This would be pretty convenient when, say indexing a particular Spatial Image with along with 1,2, or $n$ potential label masks.

Start by reviewing the existing Labels2D and SpatialImage concepts described in the issue, then determine how a channel dimension would fit the Labels representation and indexing behavior. Define what the C-equivalent labels should contain and how users access each mask; the work is done when multiple label masks can be associated with each spatial image in the proposed structure.

Written by the indexing model from the issue text.

Description

Contributor guide

First steps

Research direction

Assessment