Stereopagnosia: Fooling Stereo Networks with Adversarial Perturbations
This addresses a security and reliability problem for stereo vision systems in applications like autonomous driving, presenting a novel robustness approach specific to stereo networks.
The paper tackles the vulnerability of deep stereo networks to adversarial perturbations, showing that imperceptible image changes can significantly alter disparity maps and scene geometry, with these effects transferring across different models. It demonstrates that using such perturbations for adversarial data augmentation improves model robustness without sacrificing overall accuracy, unlike in image classification.
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.