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

Computer vision has long relied on ImageNet and other large datasets ofimages sampled from the Internet for pretraining models. However, thesedatasets have ethical and technical shortcomings, such as containing personalinformation taken without consent, unclear license usage, biases, and, in somecases, even problematic image content. On the other hand, state-of-the-artpretraining is nowadays obtained with unsupervised methods, meaning thatlabelled datasets such as ImageNet may not be necessary, or perhaps not evenoptimal, for model pretraining. We thus propose an unlabelled dataset PASS:Pictures without humAns for Self-Supervision. PASS only contains images withCC-BY license and complete attribution metadata, addressing the copyrightissue. Most importantly, it contains no images of people at all, and alsoavoids other types of images that are problematic for data protection orethics. We show that PASS can be used for pretraining with methods such asMoCo-v2, SwAV and DINO. In the transfer learning setting, it yields similardownstream performances to ImageNet pretraining even on tasks that involvehumans, such as human pose estimation. PASS does not make existing datasetsobsolete, as for instance it is insufficient for benchmarking. However, itshows that model pretraining is often possible while using safer data, and italso provides the basis for a more robust evaluation of pretraining methods.

PASS: An ImageNet replacement for self-supervised pretraining without humans