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

Self-Supervised Learning (SSL) enables training performant models usinglimited labeled data. One of the pillars underlying vision SSL is the use ofdata augmentations/perturbations of the input which do not significantly alterits semantic content. For audio and other temporal signals, augmentations arecommonly used alongside format transforms such as Fourier transforms or wavelettransforms. Unlike augmentations, format transforms do not change theinformation contained in the data; rather, they express the same information indifferent coordinates. In this paper, we study the effects of format transformsand augmentations both separately and together on vision SSL. We defineaugmentations in frequency space called Fourier Domain Augmentations (FDA) andshow that training SSL models on a combination of these and image augmentationscan improve the downstream classification accuracy by up to 1.3% onImageNet-1K. We also show improvements against SSL baselines in few-shot andtransfer learning setups using FDA. Surprisingly, we also observe that formattransforms can improve the quality of learned representations even withoutaugmentations; however, the combination of the two techniques yields betterquality.

Disentangling the Effects of Data Augmentation and Format Transform in Self-Supervised Learning of Image Representations