Point Cloud Compression Using Deep Learning

January 13, 2020 – /,

Motivation

As a developing media content, point cloud has gained great interest in both industry and research due to its applicability in diverse areas such as virtual reality, autonomous driving, and video gaming.
A high-quality 3D point cloud would enable new ways of interaction with the virtual environments. However, point clouds are highly resource-demanding and therefore are difficult to stream at an acceptable quality level.
Therefore, point cloud compression approach with a high compression ratio and tiny loss is very important to improve transportation efficiency.
In this work, the student will investigate how can compression method based on Deep learning/Auto-Encoder performing better than state-of-art compressing approaches.

Tasks

Create and evaluate point cloud compression method based on Deep learning networks. For that:

  1. Literature and related work on point cloud compression based on Deep learning/Auto-Encoder
  2. Design and implement a point cloud compression method by Deep learning
  3. Evaluation based on related work metrics such as loss, encode time, decode time and bits per point.
  4.  An Introduction for the topic, related works, design, implementation, and evaluation of the proposed system must be described and discussed in the final report

 

Requirements

  •  Good programming skills in Python
  •  Previous experience Tensorflow or Keras is very helpful


Keywords
Deep learning, Point Cloud, Compression

 

 

download corresponding tendering

Keywords: Deep learning, Point Cloud, Compression

Research Area(s):

Tutor: Alkhalili,

Open Theses