CVAT for geospatial

[AchimGaedkeLynker]

I started using CVAT for annotating aerial imagery - as alternative to QGIS.
I love the CVAT interface and the fact that it is an easily accessible browser application.

There are three observations I'd like to share and see someone has an opinion/solution:

• CVAT strips all metadata from images, esp the georeferencing
• CVAT is limited to 3 channels or one image/representation per annotation frame (e.g. I can't switch to a view highlighting water...)
• The polygon tools in QGIS are more flexible (mutliple rings/holes possible, flexible set operations) - that comes at a complexity cost though...

For geospatial data the first point is a "make or break" issue, also mentioned at #6595

The location/extent of the image are stored either as metadata in the image or as a separate "world file".
Both are not recoverable when "simply" importing a collection of tiles of aerial imagery.

I found 3 ways of dealing with it:

• Use shared storage: https://docs.cvat.ai/docs/administration/basics/installation/#share-path - so you can refer to the original file by the frame name and retrieve the necessary metadata (and original resolution/added bands)
• Use image tags to store the coordinate reference system and raster transformation
• Use an external database outside CVAT for those data

I use (at the moment) the first way to work around it...

Does anyone have another solution or a plugin I could use (or modify) for geospatial data?