Open dobkeratops opened 5 years ago
What I imagine is the graph enabling identification tasks per polygon. Eg if you annotated something as “container”, it would scan the graph nodes for that.. “what type of container is this: barrel,box,crate,cylinder,bag,...”. This could also achieve validation, ie give an option “not a container”. As the vocabulary grew over time , the resolution of descriptions could gradually increase.
that's a nice one - like that!
There are probably a few changes needed in the sourcecode to support such a mode, but I think it should be doable. :)
In order to avoid that we end up with a lot of duplicates (could happen when the same image gets served multiple times), we could try to determine the annotation coverage. If the coverage is >= 80% (just a random treshold) the image wouldn't show up anymore in this mode.
One idea I had s that an image labelling website/tool could double up somehow as reference for artists,
Recently I was looking for “2d kitbashing” (for an unrelated reason) and did indeed encounter a clip art library for 2d art... cutouts from photos:
artstation 2d kitbashing library
So imagine if someone wanting this could just flick through photos, draw around anything interesting that grabs their eye, and those can either be labelled immiediately or thrown into a queue for “identification tasks”.
But all you have you do for some sorting at the point of creation is have a vocabulary that is at least complete. Eg ... “animal,vegetable,mineral” would qualify.. just 3 labels can cover everything in the world. (But I would go a little deeper with words like container, tool, vehicle,mechanical component, structural element,...)
you could still have other people come back and refine the descriptions.. just as you are intending for people to be able to add attributes to the polygon.
The workflow for this usecase would be the exact opposite to what we have now, but all the search modes would have great utility ![Uploading 426F7239-7A8F-4BD1-AFEE-491C3DEC7C48.jpeg…]() ![Uploading AEC17BE4-EBA9-4EAE-B5E2-D431E5919BFC.jpeg…]()