Paying U-Attention to Textures: Multi-Stage Hourglass Vision Transformer for Universal Texture Synthesis

TL;DR

This paper introduces a hierarchical U-Attention vision Transformer that leverages multi-scale attention for universal texture synthesis, producing high-quality, diverse textures efficiently and generalizing well to unseen textures.

Contribution

The paper proposes a novel multi-stage hourglass vision Transformer with hierarchical attention for improved universal texture synthesis, unifying macro and micro feature attention in a single framework.

Findings

Achieves 2× better synthesis quality than previous methods.

Generalizes to unseen textures without fine-tuning.

Effective ablation results confirming component contributions.

Abstract

We present a novel U-Attention vision Transformer for universal texture synthesis. We exploit the natural long-range dependencies enabled by the attention mechanism to allow our approach to synthesize diverse textures while preserving their structures in a single inference. We propose a hierarchical hourglass backbone that attends to the global structure and performs patch mapping at varying scales in a coarse-to-fine-to-coarse stream. Completed by skip connection and convolution designs that propagate and fuse information at different scales, our hierarchical U-Attention architecture unifies attention to features from macro structures to micro details, and progressively refines synthesis results at successive stages. Our method achieves stronger 2$\times$ synthesis than previous work on both stochastic and structured textures while generalizing to unseen textures without fine-tuning. Ablation studies demonstrate the effectiveness of each component of our architecture.