Open dobkeratops opened 7 years ago
Like that! That's definitely something we could consider when starting with the scene concept.
maybe there is a possibility to gamify that?
"maybe there is a possibility to gamify that?"
it might be more in the realms of a 'pro mode', but even then the validation can still be used ("is this 'a dog looking at a bone' ").
I would hope that greater depth in what you can do with it could help keep interest in the tool.
interesting paper, https://arxiv.org/abs/1502.03044
r.e. some of the demonstrated experiments (generating textual descriptions of a scene, not just identifying what is there),
would it make sense to be able to connect object labels, and attach a label to the link . e.g.
.. that sort of thing. could be verbs, or could be prepositions ('behind', 'on', etc). it would add another layer of depth to what your dataset can represent.
(if the triple was itself another linkable object, you could also encode arbitrary sentances )