PointListNet: Deep Learning on 3D Point Lists

A sentence is a list of words. (a) text: 1D word list (b) image: 2D pixel grid (c) 3D point cloud/set (d) protein: 3D point list Figure 1. Data structure comparison of text, image, point cloud and protein. (a) Texts are regular 1D lists of words. The position is the word's sequential order in the text and the feature is the word itself. (b) Images are regular 2D grids of pixels. The position is the row and column where the pixel is located and the feature is the color. (c) Point clouds are irregular 3D point sets. The position is the 3D coordinate and the feature is the point attributes. (d) Proteins can be seen as 3D point lists. The position of an amino acid involves a regular 1D sequential order and an irregular 3D coordinate. The feature is the amino acid (residue) type.