BHViT: Binarized Hybrid Vision Transformer

TL;DR

BHViT introduces a novel binarized hybrid Vision Transformer architecture that effectively combines local information interaction, shift-based modules, and attention binarization to achieve state-of-the-art performance in energy-efficient edge deployment.

Contribution

The paper proposes BHViT, a binarization-friendly hybrid ViT architecture with new modules and techniques to improve performance of binary Vision Transformers.

Findings

Achieves state-of-the-art results among binary ViT methods.

Effectively reduces computational redundancy in token processing.

Enhances binary MLP performance with shift operation modules.

Abstract

Model binarization has made significant progress in enabling real-time and energy-efficient computation for convolutional neural networks (CNN), offering a potential solution to the deployment challenges faced by Vision Transformers (ViTs) on edge devices. However, due to the structural differences between CNN and Transformer architectures, simply applying binary CNN strategies to the ViT models will lead to a significant performance drop. To tackle this challenge, we propose BHViT, a binarization-friendly hybrid ViT architecture and its full binarization model with the guidance of three important observations. Initially, BHViT utilizes the local information interaction and hierarchical feature aggregation technique from coarse to fine levels to address redundant computations stemming from excessive tokens. Then, a novel module based on shift operations is proposed to enhance the performance of the binary Multilayer Perceptron (MLP) module without significantly increasing computational overhead. In addition, an innovative attention matrix binarization method based on quantization decomposition is proposed to evaluate the token's importance in the binarized attention matrix. Finally, we propose a regularization loss to address the inadequate optimization caused by the incompatibility between the weight oscillation in the binary layers and the Adam Optimizer. Extensive experimental results demonstrate that our proposed algorithm achieves SOTA performance among binary ViT methods.