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 is a powerful paradigm for representation learningon unlabelled images. A wealth of effective new methods based on instancematching rely on data-augmentation to drive learning, and these have reached arough agreement on an augmentation scheme that optimises popular recognitionbenchmarks. However, there is strong reason to suspect that different tasks incomputer vision require features to encode different (in)variances, andtherefore likely require different augmentation strategies. In this paper, wemeasure the invariances learned by contrastive methods and confirm that they dolearn invariance to the augmentations used and further show that thisinvariance largely transfers to related real-world changes in pose andlighting. We show that learned invariances strongly affect downstream taskperformance and confirm that different downstream tasks benefit from polaropposite (in)variances, leading to performance loss when the standardaugmentation strategy is used. Finally, we demonstrate that a simple fusion ofrepresentations with complementary invariances ensures wide transferability toall the diverse downstream tasks considered.

Why Do Self-Supervised Models Transfer? Investigating the Impact of Invariance on Downstream Tasks