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 Tools annual

Global: Zendy Free Plan

Global: Zendy Tools monthly

Global: Zendy Plus monthly

View-invariant object recognition is a challenging problem, which hasattracted much attention among the psychology, neuroscience, and computervision communities. Humans are notoriously good at it, even if some variationsare presumably more difficult to handle than others (e.g. 3D rotations). Humansare thought to solve the problem through hierarchical processing along theventral stream, which progressively extracts more and more invariant visualfeatures. This feed-forward architecture has inspired a new generation ofbio-inspired computer vision systems called deep convolutional neural networks(DCNN), which are currently the best algorithms for object recognition innatural images. Here, for the first time, we systematically compared humanfeed-forward vision and DCNNs at view-invariant object recognition using thesame images and controlling for both the kinds of transformation as well astheir magnitude. We used four object categories and images were rendered from3D computer models. In total, 89 human subjects participated in 10 experimentsin which they had to discriminate between two or four categories after rapidpresentation with backward masking. We also tested two recent DCNNs on the sametasks. We found that humans and DCNNs largely agreed on the relativedifficulties of each kind of variation: rotation in depth is by far the hardesttransformation to handle, followed by scale, then rotation in plane, andfinally position. This suggests that humans recognize objects mainly through 2Dtemplate matching, rather than by constructing 3D object models, and that DCNNsare not too unreasonable models of human feed-forward vision. Also, our resultsshow that the variation levels in rotation in depth and scale strongly modulateboth humans' and DCNNs' recognition performances. We thus argue that thesevariations should be controlled in the image datasets used in vision research.

Humans and deep networks largely agree on which kinds of variation make object recognition harder