You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
so there's a long tail of annotations that visitors wont see because they're in non-productive labels currently.
would you have time every so often to add some of the most annotated label suggestions to the productive label list (the labels that everyone sees, e.g. on the stats screen)?
it may indeed take too long to validate all of them (20,000 suggestions vs 546 productive labels currently) but adding say another 10 every so often might help
there's things like (man|woman)/(sitting|walking|running|...) .. left|right parts of other animals .. tabletop , some combinations of material ("wooden tabletop") .. new parts ("forearm") .. clothing types
it might be possible to streamline this by looking for "/" combinations of already approved words.. you might be able to just add a few more and bulk enable a few this way.
there are some mixtures of "_" and " " .. it might be possible to automatically make these synonyms ("wooden tabletop" = "wooden_tabletop") .. the rationale of _ was a grouping hint if the phrases were used in sentances or with "/" combination (e.g. "a b/c" vs "a_b/c" .. the latter is less ambiguous) .. but it hasn't been used consistently.
there's probably some good NLP neural nets around by now that could turn phrases into a useable embedding vector that could be trained on
The text was updated successfully, but these errors were encountered:
so there's a long tail of annotations that visitors wont see because they're in non-productive labels currently.
would you have time every so often to add some of the most annotated label suggestions to the productive label list (the labels that everyone sees, e.g. on the stats screen)?
it may indeed take too long to validate all of them (20,000 suggestions vs 546 productive labels currently) but adding say another 10 every so often might help
there's things like (man|woman)/(sitting|walking|running|...) .. left|right parts of other animals .. tabletop , some combinations of material ("wooden tabletop") .. new parts ("forearm") .. clothing types
it might be possible to streamline this by looking for "/" combinations of already approved words.. you might be able to just add a few more and bulk enable a few this way.
there are some mixtures of "_" and " " .. it might be possible to automatically make these synonyms ("wooden tabletop" = "wooden_tabletop") .. the rationale of _ was a grouping hint if the phrases were used in sentances or with "/" combination (e.g. "a b/c" vs "a_b/c" .. the latter is less ambiguous) .. but it hasn't been used consistently.
there's probably some good NLP neural nets around by now that could turn phrases into a useable embedding vector that could be trained on
The text was updated successfully, but these errors were encountered: