PointNetLK Revisited

We address the generalization ability of recent learningbased point cloud registration methods. Despite their success, these approaches tend to have poor performance when applied to mismatched conditions that are not wellrepresented in the training set, such as unseen object categories, different complex scenes, or unknown depth sensors. In these circumstances, it has often been better to rely on classical non-learning methods (e.g., Iterative Closest Point), which have better generalization ability. Hybrid learning methods, that use learning for predicting point correspondences and then a deterministic step for alignment, have offered some respite, but are still limited in their generalization abilities. We revisit a recent innovation-PointNetLK [1]-and show that the inclusion of an analytical Jacobian can exhibit remarkable generalization properties while reaping the inherent fidelity benefits of a learning framework. Our approach not only outperforms the stateof-the-art in mismatched conditions but also produces results competitive with current learning methods when operating on real-world test data close to the training set.