Skip to main navigation menu Skip to main content Skip to site footer

Articles

Vol. 6 No. 1 (2020): Special Issue: Proceedings of CVIS 2020

Comparison of Foveated Downsampling Techniques in Image Recognition

DOI
https://doi.org/10.15353/jcvis.v6i1.3540
Submitted
January 15, 2021
Published
2021-01-15

Abstract

Foveation is an important part of human vision, and a number of deep networks have also used foveation. However, there have been few systematic comparisons between foveating and non-foveating deep networks, and between different variable-resolution downsampling methods. Here we define several such methods, and compare their performance on ImageNet recognition with a custom Densenet network. The best variable-resolution method slightly outperforms uniform downsampling. Thus in our experiments, foveation does not substantially help or hinder object recognition in deep networks.