Machine Learning Based Point Cloud Compression
Project Description

3D sensing devices are becoming more and more popular, which generates a large amount of 3D data. Among many representation formats for 3D data, point clouds represent a good tradeoff between ease of acquisition, rendering, and processing. However, point clouds are normally represented by large amounts of data, creating big issues for storage, communication, and reconstruction. There are mature 1D/2D/3D compression techniques with their respective advantages and disadvantages. In this project, we will investigate point cloud compression with both model-driven and data-driven (machine learning) methods.

Supervisor
SONG Shenghui
Quota
1
Course type
UROP1100
UROP2100
UROP3100
UROP4100
Applicant's Roles

1. Study the current solutions and perform comparison study
2. Compare model-driven and data-driven methods
3. Develop data-driven method for point-cloud compression
4. Perform experiments for performance evaluation

Applicant's Learning Objectives

1. Understand the current solutions for point cloud compression
2. Understand model-driven and data-driven methods
3. Develop machine learning based compression method
4. Evaluate the performance of the developed method

Complexity of the project
Challenging