Twins: Revisiting the Design of Spatial Attention in Vision Transformers

TL;DR

This paper introduces simple yet effective spatial attention mechanisms in vision transformers, resulting in two architectures that outperform existing models across various visual tasks with high efficiency.

Contribution

The paper proposes two novel vision transformer architectures, Twins-PCPVT and Twins-SVT, featuring a straightforward spatial attention design that enhances performance and efficiency.

Findings

Achieved state-of-the-art results on multiple visual tasks

Demonstrated high efficiency and ease of implementation

Provided open-source code for reproducibility

Abstract

Very recently, a variety of vision transformer architectures for dense prediction tasks have been proposed and they show that the design of spatial attention is critical to their success in these tasks. In this work, we revisit the design of the spatial attention and demonstrate that a carefully-devised yet simple spatial attention mechanism performs favourably against the state-of-the-art schemes. As a result, we propose two vision transformer architectures, namely, Twins-PCPVT and Twins-SVT. Our proposed architectures are highly-efficient and easy to implement, only involving matrix multiplications that are highly optimized in modern deep learning frameworks. More importantly, the proposed architectures achieve excellent performance on a wide range of visual tasks, including image level classification as well as dense detection and segmentation. The simplicity and strong performance suggest that our proposed architectures may serve as stronger backbones for many vision tasks. Our code is released at https://github.com/Meituan-AutoML/Twins .