This repository contains a brief code review of the reference implementation of Point-GNN: Graph Neural Network for 3D Object Detection in a Point Cloud. The review was focused on studying the downsampling and graph generation algorithms.
A set of demos were developed to visulize the performance of the downsampling and graph generation algorithms. All the demos were built using the same functions that were implemented in the original Point-GNN pipeline in order to keep consistency in experimental results.