imagemonkey-core
imagemonkey-core copied to clipboard
blending for ambiguous labels, example
This is similar to the lake/river issue.. here's a scene where part is clearly a building, but another part is uncertain: it's impossible to really confirm if it's a building or just a wall,

in these cases the best thing to do IMO would be to use a blended label that says "it could be either"; it might even be possible to insert common blends into the label graph, although if their links could just be inferred from the constituents it would save doing that.. it's not always easy to figure out the name. there is certainly something walls and buildings have in common, and you can in turn say "a wall is a specialisation of the category building-or-wall", or "a building is a specialisation of the category building-or-wall".
another more specific categorisation here would be "standalone wall, or wall-of-building" .. I guess yes another way to deal with it would be to refine the "wall" label to "standalone wall" and imply "wall" could be either (tangentially, I would also like to be able to specify a blend "wall / windows" for the part on the right - you could divide a building up into it's roof and windowed surfaces: and this would be a way to bound the area containing building windows, without having to individually annotate the huge number of instances in perspective scenes)
... but how do you deal with a case like this with the task based labelling: I think this is a case where you need to be able to think about several labels in combination and how they divide up an area (similar to the man,woman,person
issue)