Google Research

Deconvolution and Checkerboard Artifacts

  • Augustus Odena
  • Vincent Dumoulin
  • Chris Olah
Distill (2016)

Abstract

When we look very closely at images generated by neural networks, we often see a strange checkerboard pattern of artifacts. These artifacts appear to be caused by deconvolutions. We demonstrate that replacing deconvolution with a "resize-convolution" causes these artifacts to disappear in a variety of contexts.

Research Areas

Learn more about how we do research

We maintain a portfolio of research projects, providing individuals and teams the freedom to emphasize specific types of work