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

Although deep-learning based video recognition models have achievedremarkable success, they are vulnerable to adversarial examples that aregenerated by adding human-imperceptible perturbations on clean video samples.As indicated in recent studies, adversarial examples are transferable, whichmakes it feasible for black-box attacks in real-world applications.Nevertheless, most existing adversarial attack methods have poortransferability when attacking other video models and transfer-based attacks onvideo models are still unexplored. To this end, we propose to boost thetransferability of video adversarial examples for black-box attacks on videorecognition models. Through extensive analysis, we discover that differentvideo recognition models rely on different discriminative temporal patterns,leading to the poor transferability of video adversarial examples. Thismotivates us to introduce a temporal translation attack method, which optimizesthe adversarial perturbations over a set of temporal translated video clips. Bygenerating adversarial examples over translated videos, the resultingadversarial examples are less sensitive to temporal patterns existed in thewhite-box model being attacked and thus can be better transferred. Extensiveexperiments on the Kinetics-400 dataset and the UCF-101 dataset demonstratethat our method can significantly boost the transferability of videoadversarial examples. For transfer-based attack against video recognitionmodels, it achieves a 61.56% average attack success rate on the Kinetics-400and 48.60% on the UCF-101. Code is available athttps://github.com/zhipeng-wei/TT.

Boosting the Transferability of Video Adversarial Examples via Temporal Translation