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

Performing inference on deep learning models for videos remains a challengedue to the large amount of computational resources required to achieve robustrecognition. An inherent property of real-world videos is the high correlationof information across frames which can translate into redundancy in eithertemporal or spatial feature maps of the models, or both. The type of redundantfeatures depends on the dynamics and type of events in the video: static videoshave more temporal redundancy while videos focusing on objects tend to havemore channel redundancy. Here we present a redundancy reduction framework,termed VA-RED$^2$, which is input-dependent. Specifically, our VA-RED$^2$framework uses an input-dependent policy to decide how many features need to becomputed for temporal and channel dimensions. To keep the capacity of theoriginal model, after fully computing the necessary features, we reconstructthe remaining redundant features from those using cheap linear operations. Welearn the adaptive policy jointly with the network weights in a differentiableway with a shared-weight mechanism, making it highly efficient. Extensiveexperiments on multiple video datasets and different visual tasks show that ourframework achieves $20\% - 40\%$ reduction in computation (FLOPs) when comparedto state-of-the-art methods without any performance loss. Project page:http://people.csail.mit.edu/bpan/va-red/.

VA-RED$^2$: Video Adaptive Redundancy Reduction