Separable Self-attention for Mobile Vision Transformers

TL;DR

This paper introduces a separable self-attention mechanism with linear complexity for mobile vision transformers, significantly improving efficiency and performance on mobile vision tasks.

Contribution

It proposes a novel separable self-attention method with linear complexity, enabling faster and more resource-efficient mobile vision transformers.

Findings

MobileViTv2 achieves 75.6% top-1 accuracy on ImageNet.

MobileViTv2 runs 3.2 times faster on mobile devices.

MobileViTv2 outperforms MobileViT by about 1% accuracy.

Abstract

Mobile vision transformers (MobileViT) can achieve state-of-the-art performance across several mobile vision tasks, including classification and detection. Though these models have fewer parameters, they have high latency as compared to convolutional neural network-based models. The main efficiency bottleneck in MobileViT is the multi-headed self-attention (MHA) in transformers, which requires $O(k^2)$ time complexity with respect to the number of tokens (or patches) $k$. Moreover, MHA requires costly operations (e.g., batch-wise matrix multiplication) for computing self-attention, impacting latency on resource-constrained devices. This paper introduces a separable self-attention method with linear complexity, i.e. $O(k)$. A simple yet effective characteristic of the proposed method is that it uses element-wise operations for computing self-attention, making it a good choice for resource-constrained devices. The improved model, MobileViTv2, is state-of-the-art on several mobile vision tasks, including ImageNet object classification and MS-COCO object detection. With about three million parameters, MobileViTv2 achieves a top-1 accuracy of 75.6% on the ImageNet dataset, outperforming MobileViT by about 1% while running $3.2\times$ faster on a mobile device. Our source code is available at: \url{https://github.com/apple/ml-cvnets}