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

Recent progress in self-supervised learning has resulted in models that arecapable of extracting rich representations from image collections withoutrequiring any explicit label supervision. However, to date the vast majority ofthese approaches have restricted themselves to training on standard benchmarkdatasets such as ImageNet. We argue that fine-grained visual categorizationproblems, such as plant and animal species classification, provide aninformative testbed for self-supervised learning. In order to facilitateprogress in this area we present two new natural world visual classificationdatasets, iNat2021 and NeWT. The former consists of 2.7M images from 10kdifferent species uploaded by users of the citizen science applicationiNaturalist. We designed the latter, NeWT, in collaboration with domain expertswith the aim of benchmarking the performance of representation learningalgorithms on a suite of challenging natural world binary classification tasksthat go beyond standard species classification. These two new datasets allow usto explore questions related to large-scale representation and transferlearning in the context of fine-grained categories. We provide a comprehensiveanalysis of feature extractors trained with and without supervision on ImageNetand iNat2021, shedding light on the strengths and weaknesses of differentlearned features across a diverse set of tasks. We find that features producedby standard supervised methods still outperform those produced byself-supervised approaches such as SimCLR. However, improved self-supervisedlearning methods are constantly being released and the iNat2021 and NeWTdatasets are a valuable resource for tracking their progress.

Benchmarking Representation Learning for Natural World Image Collections