PointNet is the official TensorFlow implementation of the landmark CVPR 2017 paper “PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation,” by Charles R. Qi, Hao Su, Kaichun Mo and Leonidas Guibas at Stanford. It introduced a neural network architecture that consumes raw point clouds directly — as unordered sets of points — instead of first converting them into voxel grids or image collections. The repository has about 5,500 stars on GitHub and remains the canonical reference implementation for one of the most influential ideas in 3D deep learning.
Point clouds from LiDAR, depth cameras and 3D scans are irregular: points have no fixed order and variable density, which breaks conventional convolutional architectures. PointNet solved this with a symmetric-function design that respects the permutation invariance of point sets, learning global and local features that can be used for both whole-shape classification and per-point semantic segmentation. The code in this repository reproduces the paper’s experiments, and its project page explains the method in more depth. Although last updated in 2023 and written for its era’s TensorFlow, the implementation is still the starting point most researchers read first, and the architecture’s ideas underpin a long line of successor methods including PointNet++.
PointNet is free and open source research code released by its authors. There is no paid tier; you run it on your own hardware and data.
PointNet is for 3D-vision researchers, robotics and autonomous-driving engineers, and students learning geometric deep learning who want the original, readable implementation of the method that made learning directly on point clouds practical.