The computers can do (many) things in our place, as illustrated by the new invention of Google. Two groups of scientists at Stanford and Google have created software that can, not to recognize an object in your photos, but to identify the whole scene and generate a picture caption in English, reports The New York Times.
This new captioning system, called Neural Legend (NIC), is based on techniques of computer vision and natural language processing. The two groups of scientists have combined neural networks “convolutional” a learning model that allowed major advances in the accuracy of computer vision, says Gigaom.
Researchers at Google say they have inspired advances in machine translation:
“[...] a recurrent neural network (RNN) transforms a sentence in French vector representation and a second RNN uses this representation to generate a target sentence in German”
Instead of first language network, the researchers used a convolutional neural network which is used to classify objects in images. One of the networks so encodait the image into a compact representation, while the other generates a sentence of description.
The system has been tested with several image databases in line. The quality of descriptive captions thus generated was judged satisfactory by an algorithm that evaluates the quality of translation between languages.
This artificial intelligence would primarily improve visual search in Online says New York Times
“The advances could enable better catalog and search billions of images and videos available online, the description and archiving are often bad. For now, the search engines like Google rely heavily on written language that accompanies a video or image to determine that it contains “
According to Gigaom, it is representative of artificial intelligences that go to more precision:
“For example, while a recognition system classical object might be able to recognize a cat and a goldfish in a picture, these new hybrid systems could probably determine that the This scene actually a cat that catches a fish in a bowl. Associated with a knowledge base that includes predator-prey relationship between the two animals, same system might be able to predict that the goldfish is about to be eaten. “
The New York Times points out that this technology could also, in the long term to help enable blind and visually impaired to understand the content of an image. Integrated robots to cars could them make better decisions in the context
But it can also affect the effectiveness of supervision.
“In the last 15 years, video cameras were placed in a large number of public and private spaces. In the future, the software that makes the camera function will not only be able to identify specific people via facial recognition, experts say, but also to identify certain types of behavior, perhaps even automatically alert authorities. “
No comments:
Post a Comment