Creating training data variability in machine learning for object labelling from images

Abstract

It is described an image labeling system () comprising: a support () for an object () to be labeled; a digital camera () configured to capture a plurality of images of a scene including said object (); a process and control apparatus () configured to receive said images and generate corresponding labeling data (-, L-L) associated to said object (); a digital display () associated with said support () and connected to the process and control apparatus () to selectively display additional images (-) selected from the group comprising: first images (-) in the form of backgrounds for the plurality of images and introducing a degree of variability in the scene; second images () indicating position and/or orientation according to which place said object () by a user on the support (); third images () to be captured by the digital camera () and provided to the process and control apparatus () to evaluate a position of the digital camera () with respect the digital display (); fourth images to be captured by the digital camera () and provided to the process and control apparatus () to evaluate at least one of the following data of the object (): position, orientation, 3D shape.