Redefine what annotation means

[corot]

Now that we can have 1..N annotations referencing a single semantic element, our previous statement that "an annotation is a semantic element in the world" doesn't hold anymore, imho.

Same way we decouple annotations and the referred data, we must decouple the definitions, e.g.

• Object | Element | Item | Dweller: a semantic element in the world (maybe object is the obvious option, but can be a bit misleading because the semantic map can contain immaterial stuff, as pickup points, room labels, etc.)
• Annotation: a descriptor of a semantic element in the world containing common properties that allow querying and graphically represent the referenced elements. An element can have 1 or more annotations.

Or go all the way around and call annotation to the data, and find another name to the descriptors (Descriptor | Handler | Pointer | Marker | Locator | ...)

[jihoonl]

https://github.com/corot/world_canvas/issues/14#issuecomment-47298277

[jihoonl]

https://github.com/corot/world_canvas/issues/14#issuecomment-47298288

[jihoonl]

https://github.com/corot/world_canvas/issues/14#issuecomment-47298304

[jihoonl]

https://github.com/corot/world_canvas/issues/14#issuecomment-47298322

[jihoonl]

https://github.com/corot/world_canvas/issues/14#issuecomment-47298342

[corot]

mmm..... your hierarchy of annotations goes well belong current capabilities of the annotations server. It only provides an aseptic "related to A, B, C..." feature with no semantics (they are given by the application)

Anyway, all this fits much better on issue #14. Here I just want to clarify world canvas vocabulary. Can I move your comments there?

[corot]

Coming back to the naming issue, thinking forward (I mean, future uses of this framework instead of existing use cases), maybe the key is to blur the distinction between annotation and data, and call annotations to semantic map elements, adding that each annotation is backed* by data in the form of a ROS message, and that more than one annotations can share this data.

Problem with this is that it sounds very artificial for existing usescases, where we already have the "backing data" that contains all the clients need, and so the annotation is somehow superfluous. But in incoming uses, the data (the ROS message) should contain just the app-specific stuff, relying on the annotation for generic stuff (pose, name, size...), claiming in favor of this naming proposal.

*This could be even optional; which backing data needs a label for a room?

[jihoonl]

Anyway, all this fits much better on issue #14. Here I just want to clarify world canvas vocabulary. Can I move your comments there?

Ah.. my bad. You are right. I moved all the comments to #14

[jihoonl]

I like Annotation - Descriptor pair. The pictures below describes descriptor and annotation relation which I have in mind. If I am still 빙글빙글 about the issue, please forgive me and lead me to right direction again. :)

Annotation-Descriptor Relation

Chair

Location

Region

[corot]

mmm.... not sure if descriptor is the right word... more than describing something with descriptor you name the object itself. I mean a descriptor must describe something, by definition, but in this case it describes nothing within the WCF. If ever, describes an object in the real world, but this is not necessary the case (e.g. elevator's waiting area, o high density trafic area)

2014-06-27 11:14 GMT+02:00 Jihoon Lee notifications@github.com:

I like Annotation - Descriptor pair. The pictures below describes descriptor and annotation relation which I have in mind. If I am still 빙글빙글 about the issue, please forgive me and lead me to right direction again. :)

Annotation-Descriptor Relation [image: untitled drawing] https://cloud.githubusercontent.com/assets/1556736/3409618/27fda4d6-fddb-11e3-9fd1-4338fd140726.png

Chair [image: copy of untitled drawing] https://cloud.githubusercontent.com/assets/1556736/3409620/2ac9249c-fddb-11e3-94a0-20620a8e62e9.png

Location [image: copy of copy of untitled drawing] https://cloud.githubusercontent.com/assets/1556736/3409621/2d0715d4-fddb-11e3-8da3-fc5a8289ecd8.png

Region [image: copy of copy of copy of untitled drawing] https://cloud.githubusercontent.com/assets/1556736/3409623/2e71b546-fddb-11e3-8bd1-54d3537a58e0.png

— Reply to this email directly or view it on GitHub https://github.com/corot/world_canvas/issues/15#issuecomment-47323367.

[stonier]

Is the semantic element really anything more than just data?

I mean it doesn't have any meaning in the world until you link it with an annotation.

annotation + data = semantic element|object?

[corot]

eeeh.... well, yes and not.

If you define annotations already as elements of the semantic world, the equation is fine. This applies to the first paragraph of my comment 8 days ago.

If you adapt existing stuff as virtual obstacles, tables, AR markers etc, that are described by a self-contained message type, the annotation is just a wrapper to make them fit on WCF. This applies to the second paragraph of my comment.

So thinking forward, my conclusion is the same that yours: annotations are semantic map elements, normally backed (but not necessarily) by data in the form of a ROS message.

[corot]

Ready to close, if everybody is happy.

[jihoonl]

:+1:

[stonier]

annotations are semantic map elements, normally backed (but not necessarily) by data in the form of a ROS message

So you're saying annotation = semantic map element?

I think it's worthwhile making the distinction between between the describing part, the data, and what would be the whole.

[stonier]

Ok...confirmed over hangout that we're happy with

semantic element = annotation + data.