Computer vision is about programming computers to see the world. It is simple to have a machine record images, cameras do that, but that is not “seeing”. By “seeing” we mean understanding what is in the images our eyes, or a camera, capture. Working out what is in a picture is an easy thing for a human to do, but for computers, computer vision is not so easy. Computer vision automates the job of “seeing” and includes analysing and understanding digital images and uses techniques such as machine learning and artificial intelligence (AI).
Fei-Fei Li is part of the team that invented what some people think was a very important project that helped to kick off recent interest and activity in AI. The project is called ImageNet. ImageNet, first presented in 2009, is now a huge database of millions of images, each image has been hand-annotated (by people) to say what objects are in the picture. The database is available for researchers to test out their image recognition software and has been used each year, since 2011, in a highly competitive challenge. The results of which give a good indication of the progress of computer vision research.
Until 2017, the data set used for the challenge was a subset of the overall ImageNet database, just 1000 images of 90 different types of dog breeds. But now things are changing, the challenge is going to be much harder and will include 3D objects. The reason for the change is that the AI are getting too good. In 2011, an AI that identified three-quarters of the images correctly was considered really good, but by 2015 AI were performing better than humans and in 2017 a child AI (an AI created by an AI) outperformed all previous ImageNet competitors.
As well as being a whizz with AI, machine learning and computer vision, Fei-Fei is passionate about access for all to these new technologies. She is a co-founder of the AI4All project which champions diversity and inclusion in AI. This educational initiative supports underrepresented groups of young people to find out about AI and aims to provide free AI learning resources too.
Science and computing: Create your own recognise an animal activity. Find, or take, photographs of lots of different animals with similar features. The feature could be eyes, noses, tails, paws. Crop the images to have just that feature and see if your friends can identify the animal. For example compare cat, dog, fox and lion noses.
Take this a step further and try using the same photographs with the machine learning for kids tools to train a machine learning model to recognise the animals and then program a game in Scratch with it.
We have lots more activities and facts about machine learning in our magazine on machine learning and dedicated webpages – but our favourite activity is brain in a bag to learn about AI and neurons with tubes and string!