SkeletonVis: Interactive Visualization for Understanding Adversarial Attacks on Human Action Recognition Models

TL;DR

SkeletonVis is an interactive visualization tool designed to help researchers understand how adversarial attacks deceive human action recognition models based on skeleton data, aiding in developing defenses.

Contribution

It introduces the first interactive system for visualizing adversarial attacks on skeleton-based human action recognition models, facilitating better understanding and defense strategies.

Findings

Enables visualization of attack mechanisms on pose models

Improves understanding of attack effects on recognition accuracy

Supports development of robust defense methods

Abstract

Skeleton-based human action recognition technologies are increasingly used in video based applications, such as home robotics, healthcare on aging population, and surveillance. However, such models are vulnerable to adversarial attacks, raising serious concerns for their use in safety-critical applications. To develop an effective defense against attacks, it is essential to understand how such attacks mislead the pose detection models into making incorrect predictions. We present SkeletonVis, the first interactive system that visualizes how the attacks work on the models to enhance human understanding of attacks.