Journée de la recherche du LTCI

L’équipe MM contribue à la journée de la recherche du LTCI avec un poster sur le dernier article publié à la conférence IEEE Multimedia Signal Processing

Version PDF du poster

 

Compression avec Deep-Learning

Les activités de recherche du groupe MM dans le domaine de la compression avec apprentissage profond ont été présentées à la journée de la recherche du département IDS


Séminaire Mariem Ben Yahia

Mercredi 4 juillet Meriem Ben Yahia exposera ses travaux sur le Streaming adaptif dynamique dans un séminaire entitulé : Dynamic Adaptive Streaming Solutions for Future Networks: HTTP/2 as a Key. Les transparents sont disponibles ici.

Best student paper award

L’article de Anas El Rhammad sur la compression des hologrammes a reçu le prix de meilleur article étudiant à la conférence SPIE Photonics Europe.

Source: Best student paper award | Site Web de Marco Cagnazzo

Nouveaux arrivants

Deux stagiaires en provenance de l’université de Naples, Chiara Scolavino et Ciro Assediato, rejoignent l’équipe pour les mois d’avril, mai et juin 2018. Vous les trouverez en salle B612. Ils travaillent sur la synthèse de vue virtuelle avec méthodes d’apprentissage profond.

Arrivée Attilio Fiandrotti

Attilio Fiandrotti a rejoint l’équipe Multimédia en qualité de Maître de Conférences en Vidéo Immersive. Son bureau est le E509. Bienvenu !

Voici quelques informations supplémentaire :

Attilio Fiandrotti received his M.Sc. and Ph.D. degrees in Computer Science from Politecnico di Torino in 2005 and 2010 respectively.
As a Ph.D. student at PoliTo and as a visiting student at EPFL, his research activities focused on robust streaming of scalable and multiview video contents over wireless networks.
Between 2010 and 2013, as a postdoctoral researcher at PoliTo, his research interests included Network Coding-based peer-to-peer video streaming and sparse signals recovery in a compressive sensing framework.
Between 2014 and 2016 he was research engineer at Sisvel group, working on a number of standardization and intellectual property related activities.
Between 2016 and 2018 he was senior researcher at Telecom Italia Joint Open Labs, researching on deep learning architectures for computer vision related applications.
Since 2018, he is with Télècom Paristech, IDS department, MM group, his research interests including deep learning techniques for immersive video analysis and compression.

Séminaires du groupe de travail ML-Compression

Des nombreux séminaires sont au programme dans le groupe de travail “ML-Compression”. Voir ici pour le programme. N’hésitez pas à vous inscrire à la liste de diffusion du groupe pour être au courant de toute nouveauté !

Groupe de travail ML + Compression

Nous avons lancé un nouveau groupe de travail et recherche sur les applications de l’apprentissage par ordinateur en compression. Ici le lien pour joindre le site du groupe.

Journée séminaires décembre 2017

Mardi 19 décembre nous accueillons quatre orateurs pour une journée de séminaires consacrée à la vidéo immersive.
Date : 19/12/2017
Lieu : salle B567, site Barrault

Agenda :
10h-11h15 : Andrei Purica (LTCI) : View synthesis and reconstruction from compressed video

11h15-12h30 J. Li (Polytech’Nantes) : Quality of Experience in Immersive Multimedia: Challenges, Methods and Perspectives

13.30-14.45 C. Ozcinar (Trinity College Dublin) : Immersive Virtual Reality Media Communication

14.45-16.00 A. Fiandrotti (Politecnico Turin) :Immersive video communications: a perspective from the network edge

Abstracts :
1)View synthesis and reconstruction from compressed video

Following the recent “boom” in connectivity, videos became the most demanded form of multimedia, with recent studies from Cisco showing they accounted for 64% of all internet traffic in 2014, with a predicted 80% by 2020. This high demand also fueled a continuous evolution of display, transmission and compression technologies which created a situation where the end used can easily access video content from a plethora of devices with various resolutions. Furthermore, the same video sequence can be usually found with different resolutions and compression levels on various cloud multimedia service providers. In addition to evolving existing technologies, there is also a lot of interest in finding the best way to provide a so-called immersive multimedia experience. Several solutions were investigated over the past years and the Multi-View video plus Depth format was found to provide a promising solution in combination with view synthesis algorithms. In this presentation I will discuss several new approaches in view synthesis and view reconstruction and show how they are used in immersive and 2D video compression and transmission systems. First, I explore the use of temporal correlations in combination with the traditional Depth-Image-Based-Rendering techniques and propose several approaches to tackle common problems in DIBR type algorithms which are shown to improve the quality of the synthesis. I also investigate the problem of multi-source video reconstruction and propose a model based framework that uses primal dual splitting proximal convex optimization algorithms to enhance the quality and resolution of videos from multiple sources with possibly different resolutions and compression levels. A discussion on the emerging 3D 360 video formats and the relevance of synthesis methods in this scenario concludes the presentation.

2) Quality of Experience in Immersive Multimedia: Challenges, Methods and Perspectives

Immersive multimedia is aiming to improve people’s viewing experience, seeking for better immersiveness or naturalness. The development of 3DTV, Virtual Reality (VR) and Augmented Reality (AR) are recent illustrative examples of this trend. The Quality of Experience (QoE) in immersive multimedia encompasses multiple perceptual dimensions.
For instance, in 3DTV, three primary dimensions have been identified: image quality, depth quality and visual comfort. In VR/AR, the dynamic viewing and interactions with virtual world and real world are new ingredients. In this talk, focusing on the most advanced immersive multimedia technology, one basic questions about QoE is studied: “how to subjectively assess QoE reliably taking care of its multidimensional aspect?” This talk will show you why the traditional standardised 2D subjective quality assessment methods do not work anymore, and what are the possible solutions. Some applications of the proposed methodology on QoE of immersive multimedia are introduced and some interesting research directions are discussed.

3)Immersive Virtual Reality Media Communication

TDB

4) Immersive video communications: a perspective from the network edge

Recent advances in image acquisition, communication and display are fostering a revamped interest for immersive video communications. On the source side, multicamera-equipped hadheld devices and omnidirectional camcorders may be the key towards many-degrees of freedom image acquisition. On the receiver side, HDR and lighfield technologies may enable high quality, eyestrain-free, realistic image display. Existing xDSL and the future 5G standards will provide the required bandwidth-rich, low-latency, channels to residential and roaming users.
Regardless of the specific technologies that will fuel the immersive communications of tomorrow, the practical implementation of such scenario over the cloud-centric, centralized, Internet of today poses a number of issues. Increased bandwidth requirements for multiview video delivery will stretch the capabilities of core distribution networks. Computational complexity of, e.g., generating virtual perspectives may exceed the computational resources available at the receiver, other than being suboptimal from a power efficiency perspective.
In order to tackle the above issues, an appealing solution is to relocate storage and computational capabilities from the network core and from user terminals towards the network edge. Future 5G networks will likely boast unprecedented cell density, offering the opportunity of pervasively caching bandwidth-intensive contents at the base stations. At the same time, recent advances in embedded parallel computing will enable offloading a number of image processing tasks to the network edge.
This presentation will briefly outline some of the potentials of network-coding based video delivery and deep learning-based image processing for video caching and processing at the network edge, respectively, discussing some of the issues to be addressed and presenting some example applications.

Journée séminaires

Pour la journée séminaire du groupe Multimédia de novembre 2017 nous accueillons deux doctorants co-encadrés avec l’IRT B<>COM de Rennes :

Nour HOBLOSS : Synthèse de vues intermédiaires dans la navigation libre de scènes 3D
Anas EL RHAMMAD : Compression d’hologrammes numériques par matching pursuit vectoriel

RDV le 17 novembre à 10h en salle B500