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

Despite impressive empirical advances of SSL in solving various tasks, theproblem of understanding and characterizing SSL representations learned frominput data remains relatively under-explored. We provide a comparative analysisof how the representations produced by SSL models differ when masking parts ofthe input. Specifically, we considered state-of-the-art SSL pretrained models,such as DINOv2, MAE, and SwaV, and analyzed changes at the representationlevels across 4 Image Classification datasets. First, we generate variations ofthe datasets by applying foreground and background segmentation. Then, weconduct statistical analysis using Canonical Correlation Analysis (CCA) andCentered Kernel Alignment (CKA) to evaluate the robustness of therepresentations learned in SSL models. Empirically, we show that not all modelslead to representations that separate foreground, background, and completeimages. Furthermore, we test different masking strategies by occluding thecenter regions of the images to address cases where foreground and backgroundare difficult. For example, the DTD dataset that focuses on texture ratherspecific objects.

Evaluating The Robustness of Self-Supervised Representations to Background/Foreground Removal