AI-assisted annotation of 3D sparse data

Abstract

Develop a method of annotating 3d sparse data (point cloud) in an efficient way with the help of deep neural network models and user corrections. Take the approach of human-in-the-loop to refine a AI generated fine annotation of the data.

Focus on the task of self-driving cars and lidar sensor observations. The model generates a denser representation of the data and refines it by leveraging interactive human 2d annotations.