How can we capture and transmit all the visual information of a scene efficiently enough for interactive use? This question is at the heart of the doctoral research of Davi Rabbouni de Carvalho Freitas, who has worked as an Early-State Researcher in the Plenoptima European Training network focusing on the exploration of the plenoptic function and its role in representing volumetric 3D data.
The study presents innovations in two main plenoptic scene representation approaches: Plenoptic Point Clouds (PPC) and Radiance Fields (RF). These techniques, while powerful in theory, face major challenges in practice, particularly in their large data sizes and computational costs, which hinder real-time rendering and streaming applications.
To tackle these issues, the dissertation proposes several contributions, consisting of a collection of four papers. First, it integrates plenoptic data into MPEG’s geometry-based encoder (G-PCC) using a hybrid compression method combining color transforms and predictive attribute coding. Next, it reduces the data footprint of real-time Neural Radiance Fields (NeRF) models by introducing a novel compression pipeline into the PlenOctrees format.
Additionally, the research proposes geometry-aware pruning techniques to compress RF models more efficiently by leveraging their structural disposition of the 3D scenes. The work also introduces a method to systematically generate and benchmark PPCs against traditional RFs using industry-standard rendering metrics.
“These techniques open the door for more responsive and immersive applications in virtual and augmented reality, remote collaboration, and digital twins,” says Freitas.
The results pave the way for advanced interactive content delivery systems where users can explore rich 3D environments with reduced latency and bandwidth demands. The research also supports the development of standardized formats for plenoptic media, ensuring wider accessibility and interoperability.
Freitas is originally from Rio de Janeiro, Brazil, and currently resides in Rennes, France. He is eager to continue applying his expertise to meaningful and practical challenges.
Public defence on Wednesday 21 May
The doctoral dissertation of MSc Davi Rabbouni Freitas in the field of plenoptic imaging, titled “Sensing and Reconstruction of Plenoptic Point Clouds”, will be publicly examined at the Centre INRIA de l’Université de Rennes on Wednesday May 21st at 15 o’clock (Europe/Helsinki), at the Aurigny room, on 263 Général Leclerc avenue, 35700, Rennes, France.
The pre-examiners were Prof. Adrian Munteanu, Vrije Universiteit Brussel, Belgium and Prof. Gauthier Lafruit, Université Libre de Bruxelles. Belgium. The opponents will be Dr. Marc Antonini, CNRS I3S, Nice-Sophia Antipolis, France and Prof. Mylène Farias, Texas State University, United States. The president will be Prof. Luce Morin, INSA Rennes, France.
The doctoral dissertation is available online.
The public defence can be followed via remote connection.
Organiser
University of Rennes, Tampere University