Self-supervised Adversarial Training of Monocular Depth Estimation against Physical-World Attacks

TL;DR

This paper presents a self-supervised adversarial training method for monocular depth estimation that enhances robustness against physical-world attacks without requiring ground-truth depth data.

Contribution

It introduces a novel self-supervised adversarial training approach leveraging view synthesis and L0-norm perturbations, specifically designed for MDE models.

Findings

Improved robustness of MDE models against adversarial attacks.

Minimal impact on the performance of benign MDE tasks.

Effective against both supervised and contrastive learning-based MDE methods.

Abstract

Monocular Depth Estimation (MDE) plays a vital role in applications such as autonomous driving. However, various attacks target MDE models, with physical attacks posing significant threats to system security. Traditional adversarial training methods, which require ground-truth labels, are not directly applicable to MDE models that lack ground-truth depth. Some self-supervised model hardening techniques (e.g., contrastive learning) overlook the domain knowledge of MDE, resulting in suboptimal performance. In this work, we introduce a novel self-supervised adversarial training approach for MDE models, leveraging view synthesis without the need for ground-truth depth. We enhance adversarial robustness against real-world attacks by incorporating L_0-norm-bounded perturbation during training. We evaluate our method against supervised learning-based and contrastive learning-based approaches specifically designed for MDE. Our experiments with two representative MDE networks demonstrate improved robustness against various adversarial attacks, with minimal impact on benign performance.