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

Recognizing objects and scenes are two challenging but essential tasks inimage understanding. In particular, the use of RGB-D sensors in handling thesetasks has emerged as an important area of focus for better visualunderstanding. Meanwhile, deep neural networks, specifically convolutionalneural networks (CNNs), have become widespread and have been applied to manyvisual tasks by replacing hand-crafted features with effective deep features.However, it is an open problem how to exploit deep features from a multi-layerCNN model effectively. In this paper, we propose a novel two-stage frameworkthat extracts discriminative feature representations from multi-modal RGB-Dimages for object and scene recognition tasks. In the first stage, a pretrainedCNN model has been employed as a backbone to extract visual features atmultiple levels. The second stage maps these features into high levelrepresentations with a fully randomized structure of recursive neural networks(RNNs) efficiently. To cope with the high dimensionality of CNN activations, arandom weighted pooling scheme has been proposed by extending the idea ofrandomness in RNNs. Multi-modal fusion has been performed through a soft votingapproach by computing weights based on individual recognition confidences (i.e.SVM scores) of RGB and depth streams separately. This produces consistent classlabel estimation in final RGB-D classification performance. Extensiveexperiments verify that fully randomized structure in RNN stage encodes CNNactivations to discriminative solid features successfully. Comparativeexperimental results on the popular Washington RGB-D Object and SUN RGB-D Scenedatasets show that the proposed approach achieves superior or on-parperformance compared to state-of-the-art methods both in object and scenerecognition tasks. Code is available athttps://github.com/acaglayan/CNN_randRNN.

When CNNs Meet Random RNNs: Towards Multi-Level Analysis for RGB-D Object and Scene Recognition