Robust quantization improves self-supervised action recognition models
QCLR: A Plug-in Robustification Module via Quantization for Contrastive Learning in Self-Supervised Skeleton-Based Human Action Recognition
Computer Vision and Pattern Recognition
Summary
Skeleton-based systems look at human movements to recognize actions, but they can be tricked by small changes meant to confuse them. The authors introduce a method called QCLR that uses quantization—a way to group data into bins—to make these systems harder to fool. This approach stops small confusing changes from affecting the recognition and keeps the systems accurate. Tests showed this method helps protect the models from tricky attacks without hurting their ability to understand actions.
What this means in practice
- •For motion capture system developers: Enhance robustness of self-supervised action recognition models against adversarial manipulations without losing accuracy.
- •For security teams in human-computer interaction: Protect action recognition systems from both known and unknown adversarial inputs using quantization techniques.
Authors
Nanzhu Chen, Feixiang He, Yunfeng Diao, Duygu Sarikaya, He Wang
Abstract
Skeleton-based human action recognition has gained increasing attention due to the stability of skeleton data across diverse environments and advances in motion capture technologies. Following the success of supervised learning approaches, self-supervised learning approaches, particularly contrastive learning ones, have achieved impressive performance. However, they remain vulnerable to adversarial attacks. In this work, we propose a plug-in robustification module via quantization for contrastive learning in self-supervised skeleton-based human action recognition (QCLR), which enhances the robustness of representations against adversarial attacks while preserving their representational capability. Specifically, we employ a dual-level quantization scheme across the encoding pipeline, which breaks the gradient propagation and absorbs small adversarial perturbations within the quantization bins. To mitigate the impact of inevitable quantization error, we further design an attention-guided multi-faceted quantization strategy to maintain the representation quality. Experiments on several benchmark datasets demonstrate that integrating QCLR into a progressive lineage of mainstream contrastive learning models substantially improves robustness against both gradient-based white-box and gradient-free black-box adversarial attacks, while preserving highly competitive classification accuracy.