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/

Zendy Plus

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

Zendy Tools

Zendy Open

Discriminative models for object classification typically learn image-basedrepresentations that do not capture the compositional and 3D nature of objects.In this work, we show that explicitly integrating 3D compositional objectrepresentations into deep networks for image classification leads to a largelyenhanced generalization in out-of-distribution scenarios. In particular, weintroduce a novel architecture, referred to as NOVUM, that consists of afeature extractor and a neural object volume for every target object class.Each neural object volume is a composition of 3D Gaussians that emit featurevectors. This compositional object representation allows for a highly robustand fast estimation of the object class by independently matching the featuresof the 3D Gaussians of each category to features extracted from an input image.Additionally, the object pose can be estimated via inverse rendering of thecorresponding neural object volume. To enable the classification of objects,the neural features at each 3D Gaussian are trained discriminatively to bedistinct from (i) the features of 3D Gaussians in other categories, (ii)features of other 3D Gaussians of the same object, and (iii) the backgroundfeatures. Our experiments show that NOVUM offers intriguing advantages overstandard architectures due to the 3D compositional structure of the objectrepresentation, namely: (1) An exceptional robustness across a spectrum ofreal-world and synthetic out-of-distribution shifts and (2) an enhanced humaninterpretability compared to standard models, all while maintaining real-timeinference and a competitive accuracy on in-distribution data.

NOVUM: Neural Object Volumes for Robust Object Classification