Google Code-in

Experiment with a visual recognition tool (http://demo.caffe.berkeleyvision.org/), with a variety of images taken from news and TV programs (you can either search for images online or take screenshots from programs which you have access to).

Experiment with around 50 images (of different types), and compare results for them. What kind of images is the tool able to produce sensible labels for? When does it get confused?

What could be the real world applications for a visual recognition tool like this?

This task is meant for you to get a very preliminary understanding of machine learning and computer vision, which are fields that are going to be very influential in the tech world in the next few years.