3DFeat-Net: Weakly Supervised Local 3D Features for Point Cloud Registration
This addresses the challenge of point cloud matching for robotics or mapping applications, but it is incremental as it builds on existing weakly supervised methods with new mechanisms.
The paper tackles the problem of learning 3D feature detectors and descriptors for point cloud registration without manual annotation, using weak supervision from GPS/INS data, and achieves state-of-the-art performance on outdoor Lidar datasets.
In this paper, we propose the 3DFeat-Net which learns both 3D feature detector and descriptor for point cloud matching using weak supervision. Unlike many existing works, we do not require manual annotation of matching point clusters. Instead, we leverage on alignment and attention mechanisms to learn feature correspondences from GPS/INS tagged 3D point clouds without explicitly specifying them. We create training and benchmark outdoor Lidar datasets, and experiments show that 3DFeat-Net obtains state-of-the-art performance on these gravity-aligned datasets.