English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Self-attention-based vision transformers (ViTs) have emerged as a highlycompetitive architecture in computer vision. Unlike convolutional neuralnetworks (CNNs), ViTs are capable of global information sharing. With thedevelopment of various structures of ViTs, ViTs are increasingly advantageousfor many vision tasks. However, the quadratic complexity of self-attentionrenders ViTs computationally intensive, and their lack of inductive biases oflocality and translation equivariance demands larger model sizes compared toCNNs to effectively learn visual features. In this paper, we propose alight-weight and efficient vision transformer model called DualToken-ViT thatleverages the advantages of CNNs and ViTs. DualToken-ViT effectively fuses thetoken with local information obtained by convolution-based structure and thetoken with global information obtained by self-attention-based structure toachieve an efficient attention structure. In addition, we use position-awareglobal tokens throughout all stages to enrich the global information, whichfurther strengthening the effect of DualToken-ViT. Position-aware global tokensalso contain the position information of the image, which makes our modelbetter for vision tasks. We conducted extensive experiments on imageclassification, object detection and semantic segmentation tasks to demonstratethe effectiveness of DualToken-ViT. On the ImageNet-1K dataset, our models ofdifferent scales achieve accuracies of 75.4% and 79.4% with only 0.5G and 1.0GFLOPs, respectively, and our model with 1.0G FLOPs outperforms LightViT-T usingglobal tokens by 0.7%.

DualToken-ViT: Position-aware Efficient Vision Transformer with Dual Token Fusion