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

Efficiently modeling dynamic motion information in videos is crucial foraction recognition task. Most state-of-the-art methods heavily rely on denseoptical flow as motion representation. Although combining optical flow with RGBframes as input can achieve excellent recognition performance, the optical flowextraction is very time-consuming. This undoubtably will count againstreal-time action recognition. In this paper, we shed light on fast actionrecognition by lifting the reliance on optical flow. Our motivation lies in theobservation that small displacements of motion boundaries are the most criticalingredients for distinguishing actions, so we design a novel motion cue calledPersistence of Appearance (PA). In contrast to optical flow, our PA focusesmore on distilling the motion information at boundaries. Also, it is moreefficient by only accumulating pixel-wise differences in feature space, insteadof using exhaustive patch-wise search of all the possible motion vectors. OurPA is over 1000x faster (8196fps vs. 8fps) than conventional optical flow interms of motion modeling speed. To further aggregate the short-term dynamics inPA to long-term dynamics, we also devise a global temporal fusion strategycalled Various-timescale Aggregation Pooling (VAP) that can adaptively modellong-range temporal relationships across various timescales. We finallyincorporate the proposed PA and VAP to form a unified framework calledPersistent Appearance Network (PAN) with strong temporal modeling ability.Extensive experiments on six challenging action recognition benchmarks verifythat our PAN outperforms recent state-of-the-art methods at low FLOPs. Codesand models are available at: https://github.com/zhang-can/PAN-PyTorch.

PAN: Towards Fast Action Recognition via Learning Persistence of Appearance