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

Feb 2019

tl;dr: First model to consume orderless list of point (point cloud raw data) for classification and semantic segmentation.

Overall impression

This paper is brilliant in creating a model that directly learns from orderless set. The use of symmetric function of max-pooling is simple, effective and leads to robust performance against outlier and missing data. The theoretical explanation and visualization of this paper is superb. This is perhaps one of the thought-provoking papers I have ever read.

Key ideas

Technical details


Notes on code