SwiftFormer: Efficient Additive Attention for Transformer-based Real-time Mobile Vision Applications

TL;DR

SwiftFormer introduces an efficient additive attention mechanism that replaces quadratic matrix operations with linear element-wise multiplications, enabling high-accuracy, real-time vision models suitable for mobile devices.

Contribution

The paper proposes a novel additive attention mechanism that reduces computational complexity and can be used throughout the network, improving speed and accuracy for mobile vision applications.

Findings

Achieves 78.5% top-1 ImageNet accuracy on small model

Runs at 0.8 ms latency on iPhone 14

Outperforms MobileViT-v2 in speed and accuracy

Abstract

Self-attention has become a defacto choice for capturing global context in various vision applications. However, its quadratic computational complexity with respect to image resolution limits its use in real-time applications, especially for deployment on resource-constrained mobile devices. Although hybrid approaches have been proposed to combine the advantages of convolutions and self-attention for a better speed-accuracy trade-off, the expensive matrix multiplication operations in self-attention remain a bottleneck. In this work, we introduce a novel efficient additive attention mechanism that effectively replaces the quadratic matrix multiplication operations with linear element-wise multiplications. Our design shows that the key-value interaction can be replaced with a linear layer without sacrificing any accuracy. Unlike previous state-of-the-art methods, our efficient formulation of self-attention enables its usage at all stages of the network. Using our proposed efficient additive attention, we build a series of models called "SwiftFormer" which achieves state-of-the-art performance in terms of both accuracy and mobile inference speed. Our small variant achieves 78.5% top-1 ImageNet-1K accuracy with only 0.8 ms latency on iPhone 14, which is more accurate and 2x faster compared to MobileViT-v2. Code: https://github.com/Amshaker/SwiftFormer