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

A few years ago, the first CNN surpassed human performance on ImageNet.However, it soon became clear that machines lack robustness on more challengingtest cases, a major obstacle towards deploying machines "in the wild" andtowards obtaining better computational models of human visual perception. Herewe ask: Are we making progress in closing the gap between human and machinevision? To answer this question, we tested human observers on a broad range ofout-of-distribution (OOD) datasets, recording 85,120 psychophysical trialsacross 90 participants. We then investigated a range of promising machinelearning developments that crucially deviate from standard supervised CNNsalong three axes: objective function (self-supervised, adversarially trained,CLIP language-image training), architecture (e.g. vision transformers), anddataset size (ranging from 1M to 1B).  Our findings are threefold. (1.) The longstanding distortion robustness gapbetween humans and CNNs is closing, with the best models now exceeding humanfeedforward performance on most of the investigated OOD datasets. (2.) There isstill a substantial image-level consistency gap, meaning that humans makedifferent errors than models. In contrast, most models systematically agree intheir categorisation errors, even substantially different ones like contrastiveself-supervised vs. standard supervised models. (3.) In many cases,human-to-model consistency improves when training dataset size is increased byone to three orders of magnitude. Our results give reason for cautiousoptimism: While there is still much room for improvement, the behaviouraldifference between human and machine vision is narrowing. In order to measurefuture progress, 17 OOD datasets with image-level human behavioural data andevaluation code are provided as a toolbox and benchmark at:https://github.com/bethgelab/model-vs-human/

Partial success in closing the gap between human and machine vision