Ripple Attention for Visual Perception with Sub-quadratic Complexity

TL;DR

Ripple attention introduces a novel, efficient attention mechanism for vision transformers that preserves spatial locality in images while reducing computational complexity, enabling improved performance on visual tasks.

Contribution

The paper proposes ripple attention, a sub-quadratic, kernel-based attention mechanism with a dynamic programming algorithm for spatially-aware, efficient vision transformer modeling.

Findings

Effective on various visual tasks

Reduces attention complexity from quadratic to sub-quadratic

Maintains spatial locality in image processing

Abstract

Transformer architectures are now central to sequence modeling tasks. At its heart is the attention mechanism, which enables effective modeling of long-term dependencies in a sequence. Recently, transformers have been successfully applied in the computer vision domain, where 2D images are first segmented into patches and then treated as 1D sequences. Such linearization, however, impairs the notion of spatial locality in images, which bears important visual clues. To bridge the gap, we propose ripple attention, a sub-quadratic attention mechanism for vision transformers. Built upon the recent kernel-based efficient attention mechanisms, we design a novel dynamic programming algorithm that weights contributions of different tokens to a query with respect to their relative spatial distances in the 2D space in linear observed time. Extensive experiments and analyses demonstrate the effectiveness of ripple attention on various visual tasks.