Improved Texture Networks: Maximizing Quality and Diversity in Feed-Forward Stylization and Texture Synthesis

TLDR

Gatys et al. showed that image style can be captured by CNN filter statistics, sparking renewed interest in texture generation and stylization, but fast generator networks still lag behind optimization-based methods in visual quality and diversity. This study aims to improve fast generator networks by introducing two key enhancements. The authors replace batch normalization with instance normalization and adopt a new learning formulation that encourages unbiased sampling from the Julesz texture ensemble to boost quality and diversity.

Abstract

The recent work of Gatys et al., who characterized the style of an image by the statistics of convolutional neural network filters, ignited a renewed interest in the texture generation and image stylization problems. While their image generation technique uses a slow optimization process, recently several authors have proposed to learn generator neural networks that can produce similar outputs in one quick forward pass. While generator networks are promising, they are still inferior in visual quality and diversity compared to generation-by-optimization. In this work, we advance them in two significant ways. First, we introduce an instance normalization module to replace batch normalization with significant improvements to the quality of image stylization. Second, we improve diversity by introducing a new learning formulation that encourages generators to sample unbiasedly from the Julesz texture ensemble, which is the equivalence class of all images characterized by certain filter responses. Together, these two improvements take feed forward texture synthesis and image stylization much closer to the quality of generation-via-optimization, while retaining the speed advantage.