DuoFormer: Leveraging Hierarchical Representations by Local and Global Attention Vision Transformer

TL;DR

DuoFormer introduces a hierarchical vision transformer that combines CNN-based multi-scale feature extraction with innovative patch tokenization and scale-wise attention, significantly improving medical image classification accuracy.

Contribution

The paper presents a novel hierarchical transformer model integrating CNNs with ViTs using patch tokenization and scale-wise attention, addressing ViTs' data dependence and enhancing multi-scale learning.

Findings

Outperforms baseline models in classification accuracy

Effectively captures intra-scale and inter-scale associations

Plug-and-play design for various CNN architectures

Abstract

Despite the widespread adoption of transformers in medical applications, the exploration of multi-scale learning through transformers remains limited, while hierarchical representations are considered advantageous for computer-aided medical diagnosis. We propose a novel hierarchical transformer model that adeptly integrates the feature extraction capabilities of Convolutional Neural Networks (CNNs) with the advanced representational potential of Vision Transformers (ViTs). Addressing the lack of inductive biases and dependence on extensive training datasets in ViTs, our model employs a CNN backbone to generate hierarchical visual representations. These representations are adapted for transformer input through an innovative patch tokenization process, preserving the inherited multi-scale inductive biases. We also introduce a scale-wise attention mechanism that directly captures intra-scale and inter-scale associations. This mechanism complements patch-wise attention by enhancing spatial understanding and preserving global perception, which we refer to as local and global attention, respectively. Our model significantly outperforms baseline models in terms of classification accuracy, demonstrating its efficiency in bridging the gap between Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs). The components are designed as plug-and-play for different CNN architectures and can be adapted for multiple applications. The code is available at https://github.com/xiaoyatang/DuoFormer.git.