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

Transformers-based methods have achieved significant performance in imagederaining as they can model the non-local information which is vital forhigh-quality image reconstruction. In this paper, we find that most existingTransformers usually use all similarities of the tokens from the query-keypairs for the feature aggregation. However, if the tokens from the query aredifferent from those of the key, the self-attention values estimated from thesetokens also involve in feature aggregation, which accordingly interferes withthe clear image restoration. To overcome this problem, we propose an effectiveDeRaining network, Sparse Transformer (DRSformer) that can adaptively keep themost useful self-attention values for feature aggregation so that theaggregated features better facilitate high-quality image reconstruction.Specifically, we develop a learnable top-k selection operator to adaptivelyretain the most crucial attention scores from the keys for each query forbetter feature aggregation. Simultaneously, as the naive feed-forward networkin Transformers does not model the multi-scale information that is importantfor latent clear image restoration, we develop an effective mixed-scalefeed-forward network to generate better features for image deraining. To learnan enriched set of hybrid features, which combines local context from CNNoperators, we equip our model with mixture of experts feature compensator topresent a cooperation refinement deraining scheme. Extensive experimentalresults on the commonly used benchmarks demonstrate that the proposed methodachieves favorable performance against state-of-the-art approaches. The sourcecode and trained models are available athttps://github.com/cschenxiang/DRSformer.

Learning A Sparse Transformer Network for Effective Image Deraining