Egocentric Perception using a Biologically Inspired Software Retina Integratedwith a Deep CNN

Abstract

We presented the concept of of a software retina, capable of significant visual data reduction in combination with scale and rotation invariance, for applications in egocentric and robot vision at the first EPIC workshop in Amsterdam [9]. Our method is based on the mammalian retino-cortical transform: a mapping between a pseudo-randomly tessellated retina model (used to sample an input image) and a CNN. The aim of this first pilot study is to demonstrate a functional retina-integrated CNN implementation and this produced the following results: a network using the full retino-cortical transform yielded an F1 score of 0.80 on a test set during a 4-way classification task, while an identical network not using the proposed method yielded an F1 score of 0.86 on the same task. On a 40K node retina the method reduced the visual data bye×7, the input data to the CNN by 40% and the number of CNN training epochs by 36%. These results demonstrate the viability of our method and hint at the potential of exploiting functional traits of natural vision systems in CNNs. In addition, to the above study, we present further recent developments in porting the retina to an Apple iPhone, an implementation in CUDA C for NVIDIA GPU platforms and extensions of the retina model we have adopted.