Preprocessing network for multiple objects

This paper describes a preprocessing network for separating objects from the background. Since most recognition networks recognize only one object at a time in the image plane, a preprocessing network must be used to pass one object at a time to the recognition network. The proposed preprocessing network consists of a growing network and an attention network. The growing network separates the object under attention from the background; while the attention network provides the necessary attention information for the growing network. Experimental results show that this approach can be applied to gray-scaled images