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 Tools annual

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 Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

Masked image modeling (MIM) has become a prevalent pre-training setup forvision foundation models and attains promising performance. Despite itssuccess, existing MIM methods discard the decoder network during downstreamapplications, resulting in inconsistent representations between pre-trainingand fine-tuning and can hamper downstream task performance. In this paper, wepropose a new architecture, RevColV2, which tackles this issue by keeping theentire autoencoder architecture during both pre-training and fine-tuning. Themain body of RevColV2 contains bottom-up columns and top-down columns, betweenwhich information is reversibly propagated and gradually disentangled. Suchdesign enables our architecture with the nice property: maintainingdisentangled low-level and semantic information at the end of the network inMIM pre-training. Our experimental results suggest that a foundation model withdecoupled features can achieve competitive performance across multipledownstream vision tasks such as image classification, semantic segmentation andobject detection. For example, after intermediate fine-tuning on ImageNet-22Kdataset, RevColV2-L attains 88.4% top-1 accuracy on ImageNet-1K classificationand 58.6 mIoU on ADE20K semantic segmentation. With extra teacher and largescale dataset, RevColv2-L achieves 62.1 box AP on COCO detection and 60.4 mIoUon ADE20K semantic segmentation. Code and models are released athttps://github.com/megvii-research/RevCol

RevColV2: Exploring Disentangled Representations in Masked Image Modeling