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

Masked Image Modeling (MIM) achieves outstanding success in self-supervisedrepresentation learning. Unfortunately, MIM models typically have hugecomputational burden and slow learning process, which is an inevitable obstaclefor their industrial applications. Although the lower layers play the key rolein MIM, existing MIM models conduct reconstruction task only at the top layerof encoder. The lower layers are not explicitly guided and the interactionamong their patches is only used for calculating new activations. Consideringthe reconstruction task requires non-trivial inter-patch interactions to reasontarget signals, we apply it to multiple local layers including lower and upperlayers. Further, since the multiple layers expect to learn the information ofdifferent scales, we design local multi-scale reconstruction, where the lowerand upper layers reconstruct fine-scale and coarse-scale supervision signalsrespectively. This design not only accelerates the representation learningprocess by explicitly guiding multiple layers, but also facilitates multi-scalesemantical understanding to the input. Extensive experiments show that withsignificantly less pre-training burden, our model achieves comparable or betterperformance on classification, detection and segmentation tasks than existingMIM models.

Masked Image Modeling with Local Multi-Scale Reconstruction