Each year, Stanford University is holding a contest ImageNet large-scale visual recognition challenge (ILSVRC) among the leading research institutions and laboratories, in which participants demonstrate the latest technological advances in this field.
The contest consists of three phases: classification, the classification of the localization and detection. In the first case, the possibility of the algorithm is estimated to create the right signature to the image, localization also implies the release of the main objects in the image, similarly formulated and detection task, but there are more stringent evaluation criteria.
In this year's competition again participated team GoogLeNet - an acronym formed from the words Google and LeNet (an implementation of convolutional neural networks). Neural network with a deeply recycled architecture can quickly learn and re-train, and produce results even in the presence of a small amount of memory due to a more than ten-fold reduction in the number of parameters in comparison to most other models of computer vision.