Over the past four years, researchers from the Google are perfecting virtual neural network to recognize more accurately images and sounds. They follow models of learning for artificial intelligence, based on neurological models of living beings. The technology won the Internet when they developed a way to reinterpret these images with psychedelic results.
Google, Microsoft and Apple will leave 20 times faster browsers
The algorithms are trained receiving thousands of different images and telling them what they are. When, for example, you show a pizza to the robot in different angles and different formats, we can say that he has been trained to recognize a pizza. Thus, to show images to the computer, it will be able to tell whether or not there pizzas them, and how safe it is that.
Recognition of the object itself is made by repeating patterns of certain features of it, such as shape and size. In Facebook , a similar and simpler feature is available when the site recognizes the faces of their friends in photos and suggests that you check them.
For the image as a whole, the computer interprets it through several layers of understanding, the first being directed to the recognition of simple things like doors and leaves, and the latter used for the recognition of complex patterns, such as trees or buildings (the union of several simple patterns). With the help of the largest planet in the database, the algorithm can also make calls between objects that have something to do, like a leaf and an insect.
Understanding the pattern recognition objects and whole picture, the next step of the engineers was to ask the computer to modify and generate images using their acquired knowledge baggage. For this, they asked the computer to interpret the image and then performed modifications in the above mentioned “understanding of layers.” The simplest modificavam shapes and colors, while more complex inseriam objects already present or related to image
Here are some examples of images generated by Deep Dream:.
As you can see, simple layers as the shape of objects are now being reproduced faithfully, but complex layers as the involved scenario will still need much improvement if the goal is the reality.
In the existing Question image manipulation and video, the results are amazing. Depending on the selected manipulation (a kind of filter), the effect of what we see varies creative even scary. Below choose some examples of manipulated images.
The resource has already been applied in a video, in the movie “Fear and Loathing in Las Vegas”. The image is already psychedelic originally and turned into something surreal by Google Deep Dream. Check.
Make yourself
The best of Google Deep Dream is that you can even use the service. If you want to modify your images with the algorithm and not know anything about programming, sites like Dreamscope App (dreamscopeapp.com) solve the task, simply upload your image and choose the filter.
But if you understand programming (mostly python), simply access the source code of the project (github.com/google/deepdream) and start the work.
What is the best Google or Baidu search engine? Comment on TechTudo Forum.
No comments:
Post a Comment