Improving on Annotation Context (meta)

[abey79]

Current situation

• Although its scope is currently narrow, annotation contexts are meant to be a somewhat generic feature to attach semantics (that the viewer understands) to specific entities.
• ClassDescription is a specific annotation: that attaches connection/colour semantics to the "keypoint" component of points2d/3d
• AnnotationInfo is a (potentially generic?) low-level data structure used for storage of annotations such as ClassDescription.

Possible improvements

• Context is both loaded and vague. We should try to find a better naming. "Annotation" (without "context") might be just as good, provided that the notion of "Annotation" is properly documented in "Concepts".
• ClassDescription could have a better name as well. It pertains to keypoint ids, so its name should probably refer to that? Also, "Class" is rather loaded and best avoided in a Python SDK.
• AnnotationInfo probably is a leaking implementation detail, and should disappear altogether from the API. ClassDescription should have a better, more explicit API instead. I guess this could be addressed by HOPE.
• One specific situation is highlighted by the mp_face example, were multiple sets of connection (which should have different colours) must be configured. This currently some manual work to create a "keypoint id -> class id" mapping and multiple ClassDescription objects.
• Timeless-ness possible improvements? (e.g. Timeless data store with UI override #1493)
• Be more explicit in the UI about annotation context existance: Points with annotation context don't show information about their annotation context on hover/select #1475
• Reconcile with ECS's System concept? Consider introducing System as a concept to fully align with ECS model. #1155
• make AnnotationContext a multi-object of ClassDescriptions instead of a field #372

Bugs

• Annotation context display is unstable with multiple classes #2403
• Transforms are not applied to connection between keypoints #2402
• Annotation context doesn't correctly project when projecting 3d keypoints into 2d image #2359