Vous êtes ici : Accueil » Kiosque » Annonce


Mot de passe : 

Mot de passe oublié ?
Détails d'identification oubliés ?


1 février 2021

Quality assessment of 3D point clouds

Catégorie : Stagiaire

Lieu: L2S, CentraleSupelec, Université Paris-Saclay

Durée : 5-6 mois

Gratification : gratification standard pour stage de recherche (env. 550€ / mois)

Three-dimensional point clouds (PC) are sets of points in the 3D space, associated to attributes such as colors, reflectance, normals, etc. They are an essential data structure in several domains, including virtual and mixed reality, immersive communication and perception in autonomous vehicles. Since PC’s easily range in the millions of points and can have complex sets of attributes, efficient point cloud compression is fundamental to enable the consumption of point cloud content in practical applications. As a result, compression of PC is currently a matter of research and standardization, e.g., the Moving Picture Expert Group (MPEG) has launched a standardization initiative to compress the geometry and/or attributes of static or dynamic point clouds [1].

In this context, measuring quantitatively the visual quality of PC’s is of fundamental importance to achieve high coding gains, as well as to enable further point cloud processing such as denoising, resampling, etc. Some recently proposed point cloud objective metrics include simple point-to-point distance; the point-to-plane metric [2], which projects the point-to-point error vector along the surface normal; angular similarity [3], which uses angular similarity between tangent planes; PC-MSDM [4], which adapts SSIM to point clouds using local mean curvature; PCQM [5], which extends the previous work by including color features; and point to distribution metric [6], which uses a new type of correspondence from point to distribution (characterized by its mean and covariance). Nevertheless, so far the accuracy of these metrics has been tested only on limited datasets. In addition, most of these quality measures are not differentiable, and thus difficult to optimize and embed in learning-based compression schemes [7].


The goal of this internship is to study new point clouds quality prediction methods based on deep learning. The biggest challenge in this context is the lack of large datasets of distorted point clouds with subjective quality annotations. Indeed, producing such datasets is expensive and time consuming. In this internship we aim at studying unsupervised and self-supervised deep learning techniques, including few-shots learning [8], with the goal to learn appropriate representations for point cloud quality assessment with existing data. Specifically, the internship will include: an initial phase of literature review, which will lead to define the scope of the work and the methodologies to consider; implementation of these methods based on the available data; validation of the proposed quality prediction algorithms on multiple PC datasets.

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) and Aladine Chetouani (Laboratoire PRISME, Université d’Orléans). The location of the stage will be the Laboratoire des Signaux et Systèmes, in CentraleSupélec, but may include a remote part depending on the evolution of the sanitary conditions. At the end of the internship, the student will have gained state-of-the-art knowledge of point cloud processing and quality assessment, as well as technical skills in using and training deep neural networks over 3D data. Depending on the candidate and the available resources, a continuation of the work as a PhD thesis will be possible.


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

Aladine Chetouani, ​aladine.chetouani@univ-orleans.fr


  1. [1] S. Schwarz et al. "Emerging MPEG standards for point cloud compression." IEEE Journal on Emerging and Selected Topics in Circuits and Systems, vol. 9, no. 1, March 2019.
  2. [2] D. Tian et al., “Geometric distortion metrics for point cloud compression,” in IEEE Intl. Conf. on Image Process. (ICIP), Beijing, Sept. 2017, pp. 3460–3464, IEEE
  3. [3] E. Alexiou and T. Ebrahimi, “Point Cloud Quality Assessment Metric Based on Angular Similarity,” in IEEE Intl. Conf. on Multimedia and Expo (ICME), July2018, pp. 1–6, ISSN: 1945-788X.
  4. [4] G. Meynet, J. Digne, and G. Lavoué, “PC-MSDM: A quality metric for 3D point clouds,” in 11th Intl.Conf. on Quality of Multimedia Experience (QoMEX),June 2019, pp. 1–3, ISSN: 2472-7814, 2372-7179
  5. [5] G. Meynet et al., “PCQM: A Full-Reference Quality Metric for Colored 3D Point Clouds,” in 12th Intl.Conf. on Quality of Multimedia Experience (QoMEX2020), Athlone, Ireland, May 2020.
  6. [6] A. Javaheri et al., “Mahalanobis Based Point to Distribution Metric for Point Cloud Geometry Quality Eval-uation,”IEEE Signal Process. Lett., vol. 27, pp. 1350–1354, 2020
  7. [7] M. Quach, G. Valenzise, F. Dufaux. “Improved Deep Point Cloud Geometry Compression.” IEEE International Workshop on Multimedia Signal Processing (MMSP'2020), Sep 2020, Tampere, Finland
  8. [8] Wang, Y., Yao, Q., Kwok, J.T. and Ni, L.M., 2020. Generalizing from a few examples: A survey on few-shot learning. ACM Computing Surveys (CSUR), 53(3), pp.1-34.

Dans cette rubrique

(c) GdR 720 ISIS - CNRS - 2011-2020.