Adversarial samples for deep monocular 6D object pose estimation

Estimating object 6D pose from an RGB image is important for many real-world applications such as autonomous driving and robotic grasping, where robustness of the estimation is crucial. In this work, for the first time, we study adversarial samples that can fool state-of-the-art (SOTA) deep learning based 6D pose estimation models. In particular, we propose a Unified 6D pose estimation Attack, namely U6DA, which can successfully attack all the three main categories of models for 6D pose estimation. The key idea of our U6DA is to fool the models to predict wrong results for object shapes that are essential for correct 6D pose estimation. Specifically, we explore a transfer-based black-box attack to 6D pose estimation. By shifting the segmentation attention map away from its original position, adversarial samples are crafted. We show that such adversarial samples are not only effective for the direct 6D pose estimation models, but also able to attack the two-stage based models regardless of their robust RANSAC modules. Extensive experiments were conducted to demonstrate the effectiveness of our U6DA with large-scale public benchmarks. We also introduce a new U6DA-Linemod dataset for robustness study of the 6D pose estimation task. Our codes and dataset will be available at \url{https://github.com/cuge1995/U6DA}.