Monday, June 18, 2018

Alphabet’s DeepMind Defies Dimensions through Advancements in Computer Vision

Article Written By: Juliana Zadarko


Photo: DeepMind

Using only a few two-dimensional images, Alphabet’s DeepMind is able to construct a three-dimensional scene using deep neural networks. Researchers have found success in reconstructing virtual environments, but current algorithms and hardware cannot support the complexity of natural environments. These advancements in computer vision technology have applications extending to police forces, where crimes may soon be recreated from a few surveillance camera images, self-driving cars, and household robots.

In each experiment, the neural network DeepMind never received more than five snapshots to create its three-dimensional interpretation. To achieve the final layout, the system constructs a simplified abstraction 
of the perceived objects solely through observation and then fills in finer details. This division of labour helps increase the speed of rendering and the accuracy of the representation of soft objects such as animals and vegetables.

Most well-known for cracking the game of Go and beating one of the world’s best players in 2016, Alphabet’s DeepMind continues to make strides in technology, focusing in areas of efficiency, assistive technology, and now computer vision.

No comments:
Write comments