Artificial intelligence taught to do artificial winter
Three experts from the American company Nvidia has developed an algorithm that allows to transform the image of the solar street picture in rainy day, snow or dead of night. The algorithm will be used for training of control systems of unmanned vehicles.
The development presented at the conference of Neural Information Processing Systems (NIPS), which runs from 4 to 9 December in Los Angeles.
During the download an error has occurred.
The algorithm is based on the use of generative adversarial networks (generative adversarial network, GAN) — vidyaranya, which has worked well in solving problems on the production of graphics data. Unlike many other similar uses GAN (for example, to create terrible landscapes), this requires human intervention is much less. Among other things, this eliminates the need to find images, such as day and night, taken from one place to the neural network could analyze the differences.
“There are many ways to use [the neural network]. For example, in California rarely rains, but we would like our unmanned vehicles worked well in the rain. We can use our method to turn Sunny California to rainy and to train unmanned vehicles,” explained one of the authors, Min-Yu Liu.