Patent attributes
A user interacts with an inventory location, such as a lane on a shelf, picking items or placing items to that inventory location. A neural network (NN) is trained using a subset of reference images to determine parameters, such as node weights. An image of the user interaction is acquired and processed by the NN to determine if an item in a test image matches an item in a candidate set of items. During use, the parameters are used to dynamically instantiate an ad hoc neural network having a graph that is based on a number of items in a candidate set. The candidate set of items may comprise items within reach of a user at the time the test image was acquired, such as on a nearby shelf or tote. The NN output indicates if the test image is deemed to match one of the candidate set of items.