The winning Google Inc (NASDAQ:GOOG) image recognition technology named GoogLeNet will likely be used by the company for YouTube and self-driving cars, the company revealed in a blog post about the system.
The revelation was made by Google Inc (NASDAQ:GOOG) as it also announced that GoogLeNet won this year’s ImageNet Large-Scale Visual Recognition Challenge (ILSVRC). According to the technology giant, GoogLeNet was named “in homage to LeNet, Yann LeCun’s influential convolutional network.”
“These technological advances will enable even better image understanding on our side and 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 Christian Szegedy wrote on the blog post.
According to the Google Inc (NASDAQ:GOOG) software engineer, GoogLeNet won the challenge’s classification and detection tasks which had extra training data. The system doubled the quality on both tasks over last year’s results, he said.
Szegedy explained that ILSVRC has three tracks: classification, classification with localization and detection. In the classification track, an algorithm’s capacity to assign labels is tested. For the second track, an added layer of location should be identified by the algorithm. Detection is the hardest track, Szegedy said because photos feature more objects which are sometimes smaller.
“Superior performance in the detection challenge requires pushing beyond annotating an image with a ‘bag of labels’ — a model must be able to describe a complex scene by accurately locating and identifying many objects in it,” the Google Inc (NASDAQ:GOOG) engineer said.
Currently, self-driving cars made by the internet giant work after the company has extensively mapped the roads they travel on. Using image-recognition technology such as GoogLeNet which identify objects such as traffic lights, cars and pedestrians may help autonomous cars from the company travel roads which are not yet thoroughly mapped by the company.
According to Szegedy, he and Wei Liu, Yangqing Jia, Pierre Sermanet, Scott Reed, Drago Anguelov, Dumitru Erhan, and Andrew Rabinovich worked on GoogLeNet. To read more on how GoogLeNet works, read Szegedy’s post on the Google Research blog.
Google Inc (NASDAQ:GOOG)’s investors includes Duquesne Capital managed by Stanley Druckenmiller. The hedge fund reported 259,610 shares in the company by the end of the second quarter of 2014.