Stage « Deep Learning-based Compression of 3D Point Clouds »

Date limite de candidature : 28/02/2022
Date de début : 01/04/2022
Date de fin : 30/09/2022

Pôle : Télécoms et réseaux
Type de poste : Stage
Contact : VALENZISE Giuseppe (giuseppe.valenzise@l2s.centralesupelec.fr)

Télécharger
la fiche

Date limite de candidature : 28/02/2022
Date de début : 01/04/2022
Date de fin : 30/09/2022

 

Pôle: Télécom et reseaux

Type de poste: stage

Contact : Giuseppe VALENZISE (giuseppe.valenzise@l2s.centralesupelec.fr)

 

Location : Laboratoire L2S, CentraleSupélec, Gif-sur-yvette (91140)

Period : 6 months in spring 2022

Gross salary : approximately 600 euros/month

Application : A National Security clearance is needed, and it can require approximately 2 months.

Contacts :

Please submit your application, including a CV and the list of your academic records (exam grades), to:

Giuseppe Valenzise, giuseppe.valenzise@l2s.centralesupelec.fr

Stéphane Coulombe, stephane.coulombe@etsmtl.ca

 

Keywords:

3D point clouds, Compression, Deep Learning

 

Titre: Deep Learning-based Compression of 3D Point Clouds

 

Context: Point clouds are sets of points in 3D space represented by spatial coordinates (x, y, z) and associated attributes, such as the color and reflectance of each point. They are an essential data structure in several domains, such as virtual and mixed reality, immersive communication, perception in autonomous vehicles, etc. Since point clouds easily range in the millions of points and can have complex sets of attributes, efficient point cloud compression (PCC) is particularly relevant, and compression of point clouds is currently a matter of research and standardization (see, e.g., the MPEG G-PCC standard). Deep point cloud compression (DPCC) is a recent research avenue exploring the use of deep neural networks for point cloud compression.

 

Objectives: The goal of this internship is to study novel methods for the compression of point cloud geometry and/or attributes. Specifically, we will study how to adapt the current auto-encoder based architectures to the varying spatial density typical of point clouds. In a first stage, the intern student will review the existing state of the art on this matter and will get familiar with DPCC toolboxes. Later, we will compare some of the previously proposed architectures for DPCC (including voxel-based convolutions, point convolutions and sparse convolutions) with the goal to improve them and design new schemes for improved coding gains. Finally, the proposed solutions will be compared with the current G-PCC codec.

 

Profile and internship details: We seek candidates with good programming skills, in particular in Python. The knowledge of deep learning frameworks (Tensorflow or Pytorch) is a desirable plus. The internship will have a duration of 5-6 months (depending on the starting date), and will be supervised by Giuseppe Valenzise (Laboratoire des Signaux et Systèmes, CentraleSupélec, France) and Stéphane Coulombe (École de technologie supérieure, Montreal, Canada). The first part of the internship will be based in France. Depending on the sanitary conditions and the interests of the candidate, it will be possible to carry out the second half of the internship in ETS, Montreal.