Mobile device platform for automated visual retail product recognition

Abstract

A computer-implemented method for configuring a mobile point-of-sale application to perform automated product-recognition checkout processes. The application includes a convolutional neural network including hidden layer n positioned downstream of remaining hidden layers and downstream of an input segment of the convolutional neural network. The method includes: appending a fine classification arm operating on output from hidden layer n and having a first loss function for generating output; appending a second classification arm operating on the output from one of the remaining n hidden layers upstream from hidden layer n and having a second loss function for generating output; providing a plurality of true labels for image subset(s) and applicable to at least one product class; and training the CNN by passing the image subsets therethrough, observing corresponding output of the first and second loss functions, and iteratively adjusting values of CNN model parameters based on the loss function outputs.