Stereopagnosia: Fooling Stereo Networks with Adversarial Perturbations

TL;DR

This paper demonstrates that imperceptible adversarial perturbations can significantly distort disparity estimates in stereo vision models, transfer across models, and improve robustness when used for adversarial data augmentation.

Contribution

It introduces a method to generate transferable adversarial perturbations for stereo networks and shows their effectiveness in enhancing model robustness without losing accuracy.

Findings

Adversarial perturbations significantly alter disparity maps.

Perturbations transfer across different stereo models.

Adversarial data augmentation improves robustness without accuracy loss.

Abstract

We study the effect of adversarial perturbations of images on the estimates of disparity by deep learning models trained for stereo. We show that imperceptible additive perturbations can significantly alter the disparity map, and correspondingly the perceived geometry of the scene. These perturbations not only affect the specific model they are crafted for, but transfer to models with different architecture, trained with different loss functions. We show that, when used for adversarial data augmentation, our perturbations result in trained models that are more robust, without sacrificing overall accuracy of the model. This is unlike what has been observed in image classification, where adding the perturbed images to the training set makes the model less vulnerable to adversarial perturbations, but to the detriment of overall accuracy. We test our method using the most recent stereo networks and evaluate their performance on public benchmark datasets.