Incorporating long-range consistency in CNN-based texture generation

TL;DR

This paper introduces a modification to CNN-based texture generation that incorporates long-range structure and symmetry constraints, improving the rendering of regular textures and symmetric images, with applications in inpainting and season transfer.

Contribution

A simple modification to CNN texture representation enabling long-range structure and symmetry constraints in image generation.

Findings

Enhanced rendering of regular textures

Improved symmetry-aware image synthesis

Effective applications in inpainting and season transfer

Abstract

Gatys et al. (2015) showed that pair-wise products of features in a convolutional network are a very effective representation of image textures. We propose a simple modification to that representation which makes it possible to incorporate long-range structure into image generation, and to render images that satisfy various symmetry constraints. We show how this can greatly improve rendering of regular textures and of images that contain other kinds of symmetric structure. We also present applications to inpainting and season transfer.