Method for scaling fine-grained object recognition of consumer packaged goods

Abstract

A method is provided for identifying consumer packaged goods (CPGs). The method comprises (a) identifying a set of objects as being a set of CPGs by applying a first level of object recognition to the set of objects, wherein said set of objects has N members, wherein N≥2, and wherein applying the first level of object recognition to the set of objects includes applying a first predefined set of criteria to the object; (b) for each of the N objects, (i) performing a second level of object recognition on the object by applying a second predefined set of criteria to the object, (ii) assigning the object to one of a plurality of predefined superclasses S=[S, . . . , S], wherein j≥2, based on the results of the second level of object recognition, (iii) applying a bounding box to the object, (iv) capturing an image of the object with an image capturing device, and (v) cropping the image to the bounding box, thereby yielding a cropped image of the object; and (c) for each object in each set SεS, (i) performing a third level of object recognition on the cropped image of the object by applying a set of criteria Cto the object, and (ii) assigning the object to one of a plurality of predefined subclasses B=[B, . . . , B], wherein k≥2, based on the results of the third level of object recognition.