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

Skeleton-based action recognition models have recently been shown to bevulnerable to adversarial attacks. Compared to adversarial attacks on images,perturbations to skeletons are typically bounded to a lower dimension ofapproximately 100 per frame. This lower-dimensional setting makes it moredifficult to generate imperceptible perturbations. Existing attacks resolvethis by exploiting the temporal structure of the skeleton motion so that theperturbation dimension increases to thousands. In this paper, we show thatadversarial attacks can be performed on skeleton-based action recognitionmodels, even in a significantly low-dimensional setting without any temporalmanipulation. Specifically, we restrict the perturbations to the lengths of theskeleton's bones, which allows an adversary to manipulate only approximately 30effective dimensions. We conducted experiments on the NTU RGB+D and HDM05datasets and demonstrate that the proposed attack successfully deceived modelswith sometimes greater than 90% success rate by small perturbations.Furthermore, we discovered an interesting phenomenon: in our low-dimensionalsetting, the adversarial training with the bone length attack shares a similarproperty with data augmentation, and it not only improves the adversarialrobustness but also improves the classification accuracy on the original data.This is an interesting counterexample of the trade-off between adversarialrobustness and clean accuracy, which has been widely observed in studies onadversarial training in the high-dimensional regime.

Adversarial Bone Length Attack on Action Recognition