Lepard: Learning partial point cloud matching in rigid and deformable scenes

We present Lepard, a Learning based approach for partial point cloud matchingin rigid and deformable scenes. The key characteristics are the followingtechniques that exploit 3D positional knowledge for point cloud matching: 1) Anarchitecture that disentangles point cloud representation into feature spaceand 3D position space. 2) A position encoding method that explicitly reveals 3Drelative distance information through the dot product of vectors. 3) Arepositioning technique that modifies the crosspoint-cloud relative positions.Ablation studies demonstrate the effectiveness of the above techniques. Inrigid cases, Lepard combined with RANSAC and ICP demonstrates state-of-the-artregistration recall of 93.9% / 71.3% on the 3DMatch / 3DLoMatch. In deformablecases, Lepard achieves +27.1% / +34.8% higher non-rigid feature matching recallthan the prior art on our newly constructed 4DMatch / 4DLoMatch benchmark.