I just downloaded a collection of 60,591 clip art images, all named things like "MH900448740.jpg" (which is, incidentally, a field of flowers).

I need some sort of machine-learning-based system that I can pour these into and get out passable text tags ("field", "flowers", "nature"), so that when I actually want to find a clip art to use I have a better than 1 in 60,000 chance of finding what I want. It doesn't have to be perfect, or even particularly good, but it does have to be better than chance.

I could pour them all into something like Clarifai, but according to their pricing page, doing them all this month would cost about $219. Plus I'd have to upload all the data.

There seem to be a lot of tools you can use to ID or tag images, but they're all online startups that want money from me on a recurring basis. Ideally I would like something I can download, and potentially pay for, and then tag all my images locally.

Does anyone have any ideas? Can I, like, set up Caffe and download GoogleNet and attach it all together with a handy Python script or something? Or is the open academic stuff not actually useful for real-world image tagging tasks?