Computer Vision Counting

The ability for a computer to visually understand the world around it is a hot topic in computer sciences. Major companies are utilizing new ways to get data into computers. Examples of this are Voice — Alexa, Siri, Google Home, Visual — the familiar and creepy: Apple facial recognition. Instagram and Snapchat have entered our lives for even more personal and intimate data collection. All of these systems want to make sense out of images and sounds and map those to concepts that a computer can make use of.

Think of those pesky CAPCHA’s that make you select all the cars and signs. This is what its all for. It’s to teach computers to see things we see and understand basic visual concepts, like ‘what a human looks like’ or ‘how cars look at all angles’.

How we do it

Google has done a fantastic job of creating open source libraries that articulate a lot of the learned objects google has been perfecting for years. For instance, Google’s Machine Learning algorithm may look at tens of millions of dog images, and it helps the computer come up with some basic features that suggest an object is a dog!

So like when you look at an animal, you can quickly tell that it is a dog, not because you compare it to all of the dogs you’ve ever seen in life. But because you know that dogs generally have four legs, are furry, have a snout, have pointy ears etc.. As you see more dogs you’re mind will expand this model to handle all the new dogs you see in life. The machine is doing a very similar process…