Visual Development and the Acquisition of Binocular Disparity Sensitivities

Abstract

This paper considers the hypothesis that systems learning aspects of visual perception may benefit from the use of suitably designed developmental progressions during training. We report the results of simulations in which three different artificial neural network models were trained to detect binocular disparities in pairs of visual images. Two of the models were developmental models in the sense that the nature of their training input changed during the course of training (either a coarse-scale-to-multiscale progression or a one-scale-to-multiscale progression). The third model was a non-developmental model in the sense that its training input remained constant during the training period. The simulation results show that the two developmental models consistently outperformed the non-developmental model. We conclude that developmental sequences during training can be useful to systems learning to detect binocular disparities. The idea that developmental progressions can aid visual learning is a viable hypothesis in need of future study.