Possible input representations of a point cloud with range R and number of points N.

(A) Three projections of the 3D point cloud onto 2D planes. Requires 3 × R2 values to represent. This data representation is used for MVCNN. (B) A voxel representation. Requires R3 values to represent. (C) Hand-crafted/designed features. Requires 28 values to represent. This data representation is used for random forest. (D) A point cloud. Requires 3 × N values to represent. Where N is the number of points per blob. This data representation is used for PointNet.