Follow
One image created by Google Artificial Neural Networks
A new batch of images created by artificial neural networks is making the rounds on our screens, looking nothing short of impressionist or surreal paintings.
Artificial neural networks (ANN) are bestowing, as you might have guessed, the capacity to recognize and separate the essence of one image from all the background noise surrounding it upon machines.
Much like the human brain would work, the images are then recreated by machines starting from that one essential point they picked out in new, startling compositions that fall nothing short of dreams or hallucinations to the human mind.
Entertaining and awe-inspiring at the same time, the ANN function based on a defined array of levels that initially filter the image loop by loop. The feedback from each loop is filtered even further by each neuron individually, as it tries to distinguish the essence of the image.
Both Google and Facebook are at the heart of this Artificial Intelligence expansion. Facebook uses two ANN that produce realistic images the size of a thumbnail.
One random vector is fed into the first ANN. The image resulted is put through further filtering until the algorithms refine the image to look much more like reality.
In a quick poll, Facebook learned that there are nearly 40 percent of people who thought that the images presented were real-life shots of an object or being rather than an ANN generated image.
One statement coming from Facebook:
“Around 40 percent of the samples generated by our class conditional LAPGAN model are realistic enough to fool a human into thinking they are real images.”
Not bad considering the art images were exclusively created by machines which took the process of creating from thinking to an almost realistic output.
Compared to Facebook, Google is not looking necessarily for realistic outputs. ANN in the case of Google releases the most whimsical images starting from a real pattern. The ANN has a free hand in deciding which visual elements are worth emphasizing.
For instance, if the ANN is being fed a flimsy photo of a cloud passing in the sky, and Google’s ANN interprets the cloud as having the shape of what is was taught to be a bird, it will work with that.
It will emphasize the traits of a bird, until the final output will make anyone believe a bird was the initial source. Of course, the Van Gogh-like backgrounds don’t help much in fooling the human eye, yet the final results are impressive.
To this extent, check out what Google creates when the ANN is fed images based exclusively on background noise. The results of the ANN filtered images are one batch of intensely colored and trippy art-like pieces.
For both Google and Facebook, the ANN are one step beyond face-recognition. Artificial intelligence is taken one step-further. And most probably, as the teams of both Facebook and Google are relentlessly working on improving ANN, we will see machine-generated images that lose their trippy character and become increasingly realistic.
Image Source: laughingsquid.com
