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

Although transformer has achieved great progress on computer vision tasks,the scale variation in dense image prediction is still the key challenge. Feweffective multi-scale techniques are applied in transformer and there are twomain limitations in the current methods. On one hand, self-attention module invanilla transformer fails to sufficiently exploit the diversity of semanticinformation because of its rigid mechanism. On the other hand, it is hard tobuild attention and interaction among different levels due to the heavycomputational burden. To alleviate this problem, we first revisit multi-scaleproblem in dense prediction, verifying the significance of diverse semanticrepresentation and multi-scale interaction, and exploring the adaptation oftransformer to pyramidal structure. Inspired by these findings, we propose anovel Semantic-aware Decoupled Transformer Pyramid (SDTP) for dense imageprediction, consisting of Intra-level Semantic Promotion (ISP), Cross-levelDecoupled Interaction (CDI) and Attention Refinement Function (ARF). ISPexplores the semantic diversity in different receptive space. CDI builds theglobal attention and interaction among different levels in decoupled spacewhich also solves the problem of heavy computation. Besides, ARF is furtheradded to refine the attention in transformer. Experimental results demonstratethe validity and generality of the proposed method, which outperforms thestate-of-the-art by a significant margin in dense image prediction tasks.Furthermore, the proposed components are all plug-and-play, which can beembedded in other methods.

SDTP: Semantic-aware Decoupled Transformer Pyramid for Dense Image Prediction