A few days ago, Google released the code for Deep Dream to the public and as we speak it’s on GitHub. Eerily insane pictures with scary eyes started flying across the internet with the #DeepDream hashtag.
The pictures may look like images from a drug induced hallucination and at times, there’s a dark, scary look about them. Here’s an image of myself as seen through Deep Dream:
What is Deep Dream?
Google recently revealed that to classify and sort its images, an artificial intelligence program known as Artificial Neural Networks is used. This technology works by spotting patterns in images in order to identify them. This artificial intelligence is already being used in Google’s own image application that could automatically recognize faces and animals.
#deepdream Dali pic.twitter.com/c1xWzQnRHJ
— codementor (@codermange) July 3, 2015
This technology is used to sort, identify and categorize images online and of course, images.google.com works based on Artificial Neural Networks (AAN).
Further Understanding Deep Dream
With #DeepDream trending on the Internet, it has been assumed that machines do dream too but how true is that? Let’s try explaining this in the simplest possible terms for easy understanding.
Google’s artificial intelligence programs can recognize images based on the data it already has. It can tell a dog is a dog by comparing the dog image you put in with what it has in it’s own database. Deep Learning is a new field within machine learning and a lot of advancement has been made over the last few years. Researchers have been working hard, training neural networks with a huge number of layers and algorithms are now learning how to classify images with much more accuracy than before.
Further more, Deep Learning algorithms are trained by feeding them with a very large number of images and they’re told what’s in each of those images. For example, once the algorithm has seen a hundred dog heads about a thousand times from a hundred angles, it has been trained and when you give it an image, it should identify dog heads within the image or tell you there’s none at all in the image.
In order to know what the algorithms “see” or “think” when given new images, Google engineers released a method for visualizing what the algorithms sees and this now allows people to see exactly what the trained neural networks were “dreaming” about in new images.
The results are the psychedelic and eerily schizophrenic images we now see on the Internet.
Why all the dog faces and strange eyes?
Almost every DeepDream image online right now were created by seeing what neural networks trained on the ImageNet dataset thinks the images are. Although this dataset has different types of images within it, but there seems to be way to many dogs, chalices and so on.
That’s why the algorithm looks into the tiniest detail in your image and tries to relate it with what it thinks it is. The result can be totally nonsensical, artistic or scary.
If the dataset were to have more diverse images, the result could’ve been different. People are now creating algorithms trained to use different datasets though.
How to create your own Deep Dream Images
If you feel like turning your beautiful family photo or selfie to a haunting nightmare, you can decide to play with the original code already at GitHub. There’s a guide on how to easily install in on Windows or Mac written by Ryan Kennedy. You can read all about it here.
Don’t wanna go through all that? These site allows their machine to “dream” about your uploaded images:
- Psychic VR Lab
- DeepDreamit
- Deepdream.akkez.ru
- deepdream.pictures/static
What about videos?
You can run videos through Deep Dream, as well as GIFs. This is “Fear and Loathing in Las Vegas” as seen through DeepDream.
