Method and Apparatus for Recognizing Objects in a Weakly Supervised Learning Manner

摘要

Provided are a method for recognizing an object in a weekly supervised learning manner and an apparatus thereof. The method for recognizing an object in a weekly supervised learning manner performed by the apparatus for recognizing the object includes the following steps of: extracting a plurality of feature maps from a training target image having a classification result of an object of interest; extracting the plurality of feature maps from the training target image; creating an activation map with respect to each object of interest by accumulating the plurality of feature maps; calculating a representative value of the object of interest by considering activation values included in the activation map of the object of interest; determining an error by comparing a classification result determined by using the representative value of the object of interest with the given classification resu and updating an object recognition model based on a convolution neural network (CNN) by back-propagating the error.