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

3D convolutional networks are prevalent for video recognition. Whileachieving excellent recognition performance on standard benchmarks, theyoperate on a sequence of frames with 3D convolutions and thus arecomputationally demanding. Exploiting large variations among different videos,we introduce Ada3D, a conditional computation framework that learnsinstance-specific 3D usage policies to determine frames and convolution layersto be used in a 3D network. These policies are derived with a two-headlightweight selection network conditioned on each input video clip. Then, onlyframes and convolutions that are selected by the selection network are used inthe 3D model to generate predictions. The selection network is optimized withpolicy gradient methods to maximize a reward that encourages making correctpredictions with limited computation. We conduct experiments on three videorecognition benchmarks and demonstrate that our method achieves similaraccuracies to state-of-the-art 3D models while requiring 20%-50% lesscomputation across different datasets. We also show that learned policies aretransferable and Ada3D is compatible to different backbones and modern clipselection approaches. Our qualitative analysis indicates that our methodallocates fewer 3D convolutions and frames for "static" inputs, yet uses morefor motion-intensive clips.

2D or not 2D? Adaptive 3D Convolution Selection for Efficient Video Recognition