SkeletonVis: Interactive Visualization for Understanding Adversarial Attacks on Human Action Recognition Models
This work addresses the vulnerability of action recognition models to adversarial attacks, which is a critical issue for applications in robotics, healthcare, and surveillance, but it is incremental as it focuses on visualization rather than a new defense method.
The paper tackles the problem of understanding adversarial attacks on skeleton-based human action recognition models, which are used in safety-critical applications like healthcare and surveillance, by presenting SkeletonVis, an interactive visualization system that enhances human understanding of how these attacks mislead models into incorrect predictions.
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.