Image segmenter output messages.

[uavster]

Messages representing the action of a class-agnostic image segmenter.

Issue: #4420

[sea-bass]

Not sure what I prefer here.

I see that the Mask2DArray message has a single header for everything, but given that the underlying sensor_msgs/Image already has a header, it is easier to just have this be a moveit_studio_vision_msgs/Mask2D[] field here?

[uavster]

Yes, I was considering this change myself today. The initial goal of Mask2DArray was to support asynchronous updates via a topic that could be synchronized with other topics based on camera timestamp (e.g. for visual servoing) without the strong pre-condition of setting each and every mask image timestamp to the same value, but there's no need to have it in the action because the result will always be paired with the goal by the action mechanism. Will change!

[uavster]

Done.

[sea-bass]

Should this message also contain some string label (for classification) and a numeric confidence score for thresholding?

[uavster]

I think we can decouple segmentation from classification here and have a Mask2DWithClassification message when we have models outputting both. I've seen this in other ROS messages. What do you think?

[sea-bass]

This should be OK if we want to do it that way, but then you'd need 2 separate behaviors with 2 separate data types in the action clients to do different tasks. So that's where I might prefer one message definition, and if we don't use classification data for certain tasks then so be it -- we ignore that part of the payload.

[uavster]

I have some doubts about the coupling-cohesion implications. Let's talk about this later. I'll merge for now, and we can add changes later.

[sea-bass]

I'm OK with these changes! If you want to change anything else regarding my comment on classification data, feel free to do that, or else we can save that for later.