A new software developed by Google engineers are able to offer a suggestion caption for your photo. The program is called Neural Image Caption (neural imagery subtitling, in free translation) and its technical basis of visual computing and natural language processing, which enable machines, in a way, see and understand words.
The project carried out by Oriol Vinyals, Alexander Toshev, Samy Bengio and Dumitru Erhan was published on the website of the Cornell University Library and published in the Google blog. The generated subtitles are descriptive, as the example above. “People can summarize a complex scene in a few words without thinking twice. But it is much harder for computers,” according to the company’s blog.
One of the origins of this technology is advancing latest in machine translation (machine translation) language. A recurrent neural network can convert an application written in a language in a vector representation and then translate it into another language sentence. The project uses a recurrent neural network as described above together with a convolutional neural network was trained to analyze images, thus creating a phrase that sums up a particular scene.
The researchers’ goal is now make the most natural subtitles, less mechanical.
No comments:
Post a Comment