PPFNet: Global Context Aware Local Features for Robust 3D Point Matching.

We present PPFNet - Point Pair Feature NETwork for deeply learning a globallyinformed 3D local feature descriptor to find correspondences in unorganizedpoint clouds. PPFNet learns local descriptors on pure geometry and is highlyaware of the global context, an important cue in deep learning. Our 3Drepresentation is computed as a collection of point-pair-features combined withthe points and normals within a local vicinity. Our permutation invariantnetwork design is inspired by PointNet and sets PPFNet to be ordering-free. Asopposed to voxelization, our method is able to consume raw point clouds toexploit the full sparsity. PPFNet uses a novel \textit{N-tuple} loss andarchitecture injecting the global information naturally into the localdescriptor. It shows that context awareness also boosts the local featurerepresentation. Qualitative and quantitative evaluations of our network suggestincreased recall, improved robustness and invariance as well as a vital step inthe 3D descriptor extraction performance.