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

Recently, three dimensional (3D) convolutional neural networks (CNNs) haveemerged as dominant methods to capture spatiotemporal representations invideos, by adding to pre-existing 2D CNNs a third, temporal dimension. Such 3DCNNs, however, are anti-causal (i.e., they exploit information from both thepast and the future frames to produce feature representations, thus preventingtheir use in online settings), constrain the temporal reasoning horizon to thesize of the temporal convolution kernel, and are not temporalresolution-preserving for video sequence-to-sequence modelling, as, forinstance, in action detection. To address these serious limitations, here wepresent a new 3D CNN architecture for the causal/online processing of videos.  Namely, we propose a novel Recurrent Convolutional Network (RCN), whichrelies on recurrence to capture the temporal context across frames at eachnetwork level. Our network decomposes 3D convolutions into (1) a 2D spatialconvolution component, and (2) an additional hidden state $1\times 1$convolution, applied across time. The hidden state at any time $t$ is assumedto depend on the hidden state at $t-1$ and on the current output of the spatialconvolution component. As a result, the proposed network: (i) produces causaloutputs, (ii) provides flexible temporal reasoning, (iii) preserves temporalresolution. Our experiments on the large-scale large Kinetics and MultiThumosdatasets show that the proposed method performs comparably to anti-causal 3DCNNs, while being causal and using fewer parameters.

Recurrent Convolutions for Causal 3D CNNs