AI could be making your food look more appetizing. And make nature look mesmerizing.
Artificial intelligence can now be used to generate convincing images of a myriad of things from hamburgers, fries, and pasta to flowers and birds and pictures of animals that look more natural than the real deal.
Artists and researchers have been working with neural networks to generate images that look real. Recently, Google intern and two researchers from Google’s DeepMind division released a paper featuring AI-generated images that look mind-blowingly real.
Andrew Brock and fellow researchers at Google have created a generative adversarial network (GAN) to produce new images.GAN is a type of AI algorithms that uses two neural networks (Generators and Discriminators) that compete with each other to generate images. This particular GAN, known as the BigGAN, was initially trained on thousands of images linked to specific words.
Deep Mind’s BigGAN generates scarily accurate images: n image of a dog that easy looks like one you would find in real life, a juicy-looking burger, a shot of a bird perched on a tree.
Credits: Andrew Brock
Credits: Andrew Brock
However, while the algorithm an generate a juicy looking burger ready to be eaten with fries and coke, it still has its imperfections since neural networks depend on a large number of images and still can’t count or interpret larger structural trends. So while these images show the successfully generated realistic images, there are also examples of AI-generated images that show animals with multiple heads or tails or too many horns.