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

Object recognition and viewpoint estimation lie at the heart of visualunderstanding. Recent works suggest that convolutional neural networks (CNNs)fail to generalize to out-of-distribution (OOD) category-viewpointcombinations, ie. combinations not seen during training. In this paper, weinvestigate when and how such OOD generalization may be possible by evaluatingCNNs trained to classify both object category and 3D viewpoint on OODcombinations, and identifying the neural mechanisms that facilitate such OODgeneralization. We show that increasing the number of in-distributioncombinations (ie. data diversity) substantially improves generalization to OODcombinations, even with the same amount of training data. We compare learningcategory and viewpoint in separate and shared network architectures, andobserve starkly different trends on in-distribution and OOD combinations, ie.while shared networks are helpful in-distribution, separate networkssignificantly outperform shared ones at OOD combinations. Finally, wedemonstrate that such OOD generalization is facilitated by the neural mechanismof specialization, ie. the emergence of two types of neurons -- neuronsselective to category and invariant to viewpoint, and vice versa.

When and how CNNs generalize to out-of-distribution category-viewpoint combinations