Institute of Communications and Connected Systems


Predicting point cloud over time

Black gradating to dark blue background and floor with grid lights. Graphic illustration, outline shape only, of small groups of men and woman standing together. Foreground: Left hand side within a circle, an avatar wearing a VR style headset.

1 February 2021

Proposing an end-to-end learning network to predict future frames in a point cloud sequence

Funder N/A
Amount N/A 

Project website LASP                                    

Research topics volumetric video | point cloud prediction | mixed reality


Volumetric video (VV), which allows 3D representation of real-world scenes and objects to be visualized from any viewpoint or viewing direction, is an emergent digital media for mixed reality systems. As such, VV enables novel formats of creative storytelling, which are immersive, interactive, and include real-world content. VV also enables novel forms of immersive real-time communication also known as telepresence.
However, the higher level of immersiveness and realism offered to VV comes with many new open challenges, such as processing of point cloud content (one possible representation of VV). This has motivated intense research toward PC processing, with strong focus on static PCs, leaving the dynamic PC processing usually overlooked. In this project, we focus on dynamic PC processing and specifically on the prediction of point cloud sequence.


P. Gomes, S. Rossi, L. Toni, "Spatio-Temporal Graph-RNN for Point Cloud Prediction", IEEE ICIP 2021 https://arxiv.org/abs/2102.07482

View Principal Investigator's Publications