About the robustness of the CPnP method

[bereze]

Hi, I've seen the paper and it's a great work. I have one question about the robustness of the method——How robust is this method when facing scenes with outlier points, such as 3D-2D correspondence in SLAM? How does the CPnP method handle the outliers ? Can it effectively eliminate them?

[guangyangzeng]

Thank you for your interest in our work. We have not yet analyzed the scenes with outlier points, and in our experiments on the ETH3D dataset, no outlier point detection and removal algorithm has been implemented before the PnP estimation. The experiment result is presented in our paper.

The robustness to the scenes with outlier points is indeed a critical property of a PnP algorithm. Thank you for your nice question. In future work, we will apply our algorithm to more real datasets to test its robustness. At the same time, we will analyze the impact of outlier points theoretically and utilize some algorithms, e.g., the RANSAC method to ease their influence.