A geometrically aware auto-encoder for multi-texture synthesis

TL;DR

This paper introduces a geometrically aware auto-encoder for multi-texture synthesis that effectively disentangles texture and spatial information, outperforming existing methods in quality and metrics.

Contribution

It presents a novel auto-encoder architecture combining second order neural statistics and adaptive periodic content for improved texture synthesis.

Findings

Outperforms state-of-the-art methods in visual quality

Achieves better texture-related metrics

Enables effective texture interpolation

Abstract

We propose an auto-encoder architecture for multi-texture synthesis. The approach relies on both a compact encoder accounting for second order neural statistics and a generator incorporating adaptive periodic content. Images are embedded in a compact and geometrically consistent latent space, where the texture representation and its spatial organisation are disentangled. Texture synthesis and interpolation tasks can be performed directly from these latent codes. Our experiments demonstrate that our model outperforms state-of-the-art feed-forward methods in terms of visual quality and various texture related metrics.