I’m playing around with some image detection routines, trying to integrate them into my Great Image Categorization Project that’s been idling for ages now. I’ve been using Darknet as an engine and a few different pre-trained models, mostly YOLO9000 as of late. But these may change, as I’m getting some…interesting results.
gender inequality in image recognition: a bird’s story 
what makes a man a leader? 
woof 
how close can we get to being right while still being wrong? 
no 
actually lithuanian on her mother’s side if wikipedia is to be believed
Which isn’t to say that it’s all wrong and odd answers. Sometimes it’s pretty much dead-on.
yes 
turrible 
interesting choices here 
beautiful and he knows it 
this was an animated .gif; more on that later 
whoa
And sometimes the engine just…falls asleep and gives me nothing. This merits further investigation, but I just haven’t had time yet.
The process isn’t close to perfect, of course — which is what I expected, there’s just no way of creating a perfect, automatic image indexing solution — but it is an improvement, and that’s all I was asking of this particular project. Actually, there’s a bit more to this endeavor, but that’s for another post.