Google has shown off its winning entry in an annual computer vision challenge whose entrants include both academic institutions and industry labs, and made its work available to other researchers.
In this year’s challenge, team GoogLeNet tasks, doubling the quality on both tasks over last year’s results. The team participated with an open submission, meaning that the exact details of its approach are shared with the wider computer vision community to foster collaboration and accelerate progress in the field …
Google cites its self-driving cars as one of the obvious applications of the technology.
The progress is directly transferable to Google products such as photo search, image search, YouTube, self-driving cars, and any place where it is useful to understand what is in an image as well as where things are.
Google’s autonomous cars currently rely on extremely detailed mapping of the specific roads on which they are used, the computer system being taught to recognize each individual feature, down to individual stop signs. A system which can intelligently identify objects for itself would clearly be a much more useful approach if it can be made sufficiently fast and reliable. A large part of the GoogLeNet focus is on using approaches which are fast and make highly efficient use of computing resources.
Robotics are another obvious application of image-recognition systems.