Categories, tags, captions are inconsistent
My application is trying to determine images with people in them. I thought I would just be able to look for a certain tag or category but these do not seem to be consistent, even within the one result.
For example, I have a result with the caption "a close up of a person". It has a category of "people_", score: 0.75 and category of "outdoor_", score: 0.003. So far so good. However, it only has 1 tag of "outdoor", confidence: 0.99. There is no person/people tag (at the high level - there are description tags but without any score these are pretty useless).
This might be ok if I could look for the people_ category, but this rarely appears even for clear images of people.
To add to this point, I just saw an example where faces.gender = "Male" but description.captions = "a woman holding a tennis racket". To be clear, there was only 1 record in both faces and description arrays.