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

This paper introduces Content-aware Token Sharing (CTS), a token reductionapproach that improves the computational efficiency of semantic segmentationnetworks that use Vision Transformers (ViTs). Existing works have proposedtoken reduction approaches to improve the efficiency of ViT-based imageclassification networks, but these methods are not directly applicable tosemantic segmentation, which we address in this work. We observe that, forsemantic segmentation, multiple image patches can share a token if they containthe same semantic class, as they contain redundant information. Our approachleverages this by employing an efficient, class-agnostic policy network thatpredicts if image patches contain the same semantic class, and lets them sharea token if they do. With experiments, we explore the critical design choices ofCTS and show its effectiveness on the ADE20K, Pascal Context and Cityscapesdatasets, various ViT backbones, and different segmentation decoders. WithContent-aware Token Sharing, we are able to reduce the number of processedtokens by up to 44%, without diminishing the segmentation quality.

Content-aware Token Sharing for Efficient Semantic Segmentation with Vision Transformers