S2WAT: Image Style Transfer via Hierarchical Vision Transformer using Strips Window Attention

TL;DR

S2WAT introduces a hierarchical vision transformer with diverse window attention and adaptive merging for improved image style transfer, effectively capturing both local and global dependencies.

Contribution

The paper proposes a novel hierarchical vision transformer architecture with diverse window shapes and an adaptive attention merge strategy for enhanced style transfer.

Findings

Outperforms state-of-the-art transformer-based style transfer methods

Effectively captures both short- and long-range dependencies

Demonstrates superior visual quality on benchmark datasets

Abstract

Transformer's recent integration into style transfer leverages its proficiency in establishing long-range dependencies, albeit at the expense of attenuated local modeling. This paper introduces Strips Window Attention Transformer (S2WAT), a novel hierarchical vision transformer designed for style transfer. S2WAT employs attention computation in diverse window shapes to capture both short- and long-range dependencies. The merged dependencies utilize the "Attn Merge" strategy, which adaptively determines spatial weights based on their relevance to the target. Extensive experiments on representative datasets show the proposed method's effectiveness compared to state-of-the-art (SOTA) transformer-based and other approaches. The code and pre-trained models are available at https://github.com/AlienZhang1996/S2WAT.