PointNet: Deep Learning on Point Sets for 3D Classifification and Segmentation

• We design a novel deep net architecture suitable for consuming unordered point sets in 3D
• We show how such a net can be trained to perform 3D shape classifification, shape part segmentation and scene semantic parsing tasks

Properties of Point Sets in R n

• Unordered （所以网络要对 N！种点排列都work）
• Interaction among points （ the model needs to be able to capture local structures from nearby points, and the combinatorial interactions among local structures.
• Invariance under transformations （简而言之 就是对旋转或者平移这种变换不敏感）

solution

Symmetry Function for Unordered Input

MLP directly on the sorted point set performs poorly,

简而言之就是MPL 再加single variable function 再加maxpooling

Local and Global Information Aggregation

point segmentation requires a combination of local and global knowledge.

Joint Alignment Network

Invariance under transformations

We predict an affifine transformation matrix by a mini-network

PointNet++: Deep Hierarchical Feature Learning on Point Sets in a Metric Space