Slowly Google continues to demonstrate that much of its work teams today are immersed in a project destined to change the way in which we understand the world of technology. A very specific case is the so-called as Brain Team of Google, responsible for the ‘Show and Tell‘project team, a software where we have worked to provide an artificial intelligence system capable of adding photographs a small footer that describes the content of the image itself.
To achieve these results, you have a couple of images scattered throughout this same entry where much better understand the process carried out, to let know you that this team of engineers has used TensorFlow as the basis to be able to generate those captions. The result of all the work is, as announce their responsible, that this type of software is capad of offer to a 93.9% of precision in their analysis.
‘ Show and Tell’ is the name of a new software from Google capable of add feet of photo to images with a description of what occurs in the same.
As often happens, to achieve this accuracy, has been to train the algorithm with captions created by own components of the team Brain Team. Thanks to this, as the developers of this software indicated, has been that the descriptions are well constructed phrases and not only combinations of object names.
On the other hand, this new evolution of the system has achieved that now it is much faster. If train system in its first version it was supposed to spend three seconds per image using a GPU NVIDIA G20, now, in the new version released, training will only take us about 0.7 seconds.
More information: google
