DeepDream formed by Google is described by Wikipedia as a “convolutional neural network to discover and improve patterns in images via algorithmic pareidolia, thus generating a dreamlike hallucinogenic manifestation in the intentionally over-processed images”.
The transformation of pictures into a dreamlike image only with the use of certain algorithms has marked the next stage of evolution in not only photography and photo editing, but also technology and human genius.
The artificial neural network created by Google, known as DeepDream, to interpret imagery is not only intriguing because of its beauty but also the ingenuity of the entire idea.
How It Works:
The algorithm was initially designed to notice the edges in the images and to state what the object in the image is, such as “This is a tiger”, or “That is a bone”. It was observed that the same neural network could be used by the algorithms to process its own images, instead of the actual object in the image.
For example, if the algorithm has processed a pigeon in an image by the recognition of certain edges and angles, it will continue to process and enhance them in a manner that you can observe the pigeon in the image, even if there isn’t one present there. The DeepDream algorithm achieves this through continuous processing of the object and transforming it into a dreamlike embodiment.
Why is it so desired?
The general masses seem to be intrigued by this new development, by Google. The basic reason for such a predicament is that the images thus generated by the DeepDream process are looked at as a form of art. These images are:
1. Beautiful: Any object of beauty recognised by an individual is bound to attract some attention.
2. Out-of-the-box: The idea itself, of a dreamlike image, is something very few could have imagined. To be able to use it for individual purposes is unique and the desire to do so is almost compelling.
3. Abstract: Modern art gains its fan following due to its abstract approach and subjective interpretation. Similar is the condition with DeepDream imagery.
4. Wanted: Last, but not the least, a thing of beauty or originality cannot be made popular without advertisement or public following and support. This new technology has gained the support of the crowds from the beginning and has thus become such a hot commodity in the market.
All of these above reasons have had some major role to play in the increase in the number of users for the DeepDream algorithm program. The dreamlike images produced by it have had a significant impact on people globally.
The Science Behind it:
The artificial neural network is trained by the software engineers to recognise as well as adjust certain angles and edges in the images. Around 10-30 layers of artificial neurons are stacked together for the programming to commence. These neurons are fed millions of images and shown how to adjust the parameters to suit that of the user’s choices. Each image is fed into the input layer, which then feeds it to the next, till the desired output is attained.
The entire designing process is a long and complicated one, yet the results are outstanding and worth all the effort that is put in.
The actual level of enhancement and the amount of complexity in the images generated depends upon the layer the engineers choose to enhance it on. The network is also able to choose on its own, the layer of output or the level of enhancement. Any image thus generated is completely the computer’s own artwork, made possible by the hard work of certain software engineers.
DeepDream algorithm has marked the next age of imagery and computer intelligence developed and enhanced by human beings.
LSI Keywords: DeepDream algorithm, computer intelligence, artificial neural network, Science Behind it, intrigued by this new development, created by Google, programming, software engineers, layers of artificial neural networks, adjust certain angles and edges
References:
https://research.googleblog.com/2015/06/inceptionism-going-deeper-into-neural.html
https://artofericwayne.com/2015/07/08/google-deep-dream-getting-too-good/