Search

Recent advances in full body 3D reconstruction methods have lead to the realisation of high quality, real-time, photo realistic capture of users in a range of tele-immersion (TI) contexts including gaming and mixed reality environments. The full body reconstruction (FBR) process is computationally expensive requiring comparatively high CPU, GPU and network resources in order to maintain a shared, virtual reality in which high quality 3D reproductions of users can be rendered in real-time. A significant optimisation of the delivery of FBR content has been achieved through the realtime compression and de-compression of 3D geometry and textures. Here we present a new, adaptive compression methodology that allows a TI system called 3D-LIVE to modify the quality and speed of a FBR TI pipeline based on the data carrying capability of the network. Our rule-based adaptation strategy uses network performance sampling processes and a configurable rule engine to dynamically alter the compression of FBR reconstruction on-the-fly. We demonstrate the efficacy of the approach with an experimental evaluation of system and conclude with a discussion of future directions for adaptive FBR compression. This work was supported by the EU funded project 3DLIVE, GA 318483. http://3dliveproject.eu/

International audience; We combine a mask motion detection algorithm with both the WAAVES adaptive compression algorithm (resulting into MMWaaves) and a JPEG2000 coder (resulting into MMJPEG2000) for the compression of medical images sequences. Several images were compressed using Waaves, MMWaaves, and MMJPEG2000 to observe which tool provided the best visual quality while maintaining a high compression ratio. Compared to Waaves, the MMWaaves achieved compression gains up to 40% for CT scans and 50 % for MRI. In addition, the SSIM values attributed to the compressed images were between 0.96 and 0.988 while the PSNR values were higher than 42. In addition MMWaaves attained a superior performance than MMJPEG2000.

International audience; This paper sheds light on universal coding with respect to classes of memoryless sources over a countable alphabet defined by an envelope function with finite and non-decreasing hazard rate. We prove that the auto-censuring AC code introduced by Bontemps (2011) is adaptive with respect to the collection of such classes. The analysis builds on the tight characterization of universal redundancy rate in terms of metric entropy % of small source classes by Opper and Haussler (1997) and on a careful analysis of the performance of the AC-coding algorithm. The latter relies on non-asymptotic bounds for maxima of samples from discrete distributions with finite and non-decreasing hazard rate.

International audience; This paper deals with the problem of universal lossless coding on a countable infinite alphabet. It focuses on some classes of sources defined by an envelope condition on the marginal distribution, namely exponentially decreasing envelope classes with exponent $\alpha$. The minimax redundancy of exponentially decreasing envelope classes is proved to be equivalent to $\frac{1}{4 \alpha \log e} \log^2 n$. Then a coding strategy is proposed, with a Bayes redundancy equivalent to the maximin redundancy. At last, an adaptive algorithm is provided, whose redundancy is equivalent to the minimax redundancy

In this PhD thesis we present the results we obtained in three linked fields: data compression for infinite alphabets; infinite-dimensinal Bayesian Statistics; multivariate multinomial mixture models. The first point deals with the problem of universal lossless coding on a countable infinite alphabet. It focuses on some classes of stationary memoryless sources defined by an envelope condition on the marginal distribution, namely exponentially decreasing envelope classes. An equivalent of the minimax redundancy of such classes is obtained. Then an approximately maximin prior distribution is provided and an adaptive algorithm is proposed, whose maximum redundancy is equivalent to the minimax redundancy. The next works deal with the asymptotic normality of a-posteriori distributions (Bernstein-von Mises theorems) in several nonparametric and semiparametric frameworks. First, in Gaussian linear regression models when the number of regressors increases with the sample size. Two kinds of Bernstein-von Mises Theorems are obtained in this framework: nonparametric theorems for the parameter itself, and semiparametric theorems for functionals of the parameter. We apply them to the Gaussian sequence model and to the regression of functions in Sobolev and Hölder regularity classes, in which we get the minimax convergence rates. Adaptivity is reached for the Bayesian estimators of functionals in our applications. We also get a nonparametric Bernstein-von Mises theorem for increasing-dimensional exponential models. In the last part of our work we consider the problem of estimating the number of components and the relevant variables in a multivariate multinomial mixture, in order to perform an unsupervised classification. Such models arise in particular when dealing with multilocus genotypic data. A new penalized maximum likelihood criterion is proposed, and a non-asymptotic oracle inequality is obtained. The criterion used in practice needs a calibration thanks to the slope heuristics, in an automatic data-driven procedure. Using simulated data, we found that this procedure improves the performances of the selection procedure with respect to classical criteria such as BIC and AIC. The procedures are implemented in a free-of-charge software., Dans cette thèse de doctorat, nous présentons les travaux que nous avons effectués dans trois directions reliées : la compression de données en alphabet infini, les statistiques bayésiennes en dimension infinie, et les mélanges de distributions discrètes multivariées. Dans le cadre de la compression de données sans perte, nous nous sommes intéressé à des classes de sources stationnaires sans mémoire sur un alphabet infini, définies par une condition d'enveloppe à décroissance exponentielle sur les distributions marginales. Un équivalent de la redondance minimax de ces classes a été obtenue. Un algorithme approximativement minimax ainsi que des a-priori approximativement les moins favorables, basés sur l'a-priori de Jeffreys en alphabet fini, ont en outre été proposés. Le deuxième type de travaux porte sur la normalité asymptotique des distributions a-posteriori (théorèmes de Bernstein-von Mises) dans différents cadres non-paramétriques et semi-paramétriques. Tout d'abord, dans un cadre de régression gaussienne lorsque le nombre de régresseurs augmente avec la taille de l'échantillon. Les théorèmes non-paramétriques portent sur les coefficients de régression, tandis que les théorèmes semi-paramétriques portent sur des fonctionnelles de la fonction de régression. Dans nos applications au modèle de suites gaussiennes et à la régression de fonctions appartenant à des classe de Sobolev ou de régularité hölderiennes, nous obtenons simultanément le théorème de Bernstein-von Mises et la vitesse d'estimation fréquentiste minimax. L'adaptativité est atteinte pour l'estimation de fonctionnelles dans ces applications. Par ailleurs nous présentons également un théorème de Bernstein-von Mises non-paramétrique pour des modèles exponentiels de dimension croissante. Enfin, le dernier volet de ce travail porte sur l'estimation du nombre de composantes et des variables pertinentes dans des modèles de mélange de lois multinomiales multivariées, dans une optique de classification non supervisée. Ce type de modèles est utilisé par exemple pour traiter des données génotypiques. Un critère du maximum de vraisemblance pénalisé est proposé, et une inégalité oracle non-asymptotique est obtenue. Le critère retenu en pratique comporte une calibration grâce à l'heuristique de pente. Ses performances sont meilleurs que celles des critères classiques BIC et AIC sur des données simulées. L'ensemble des procédures est implémenté dans un logiciel librement accessible.

International audience; We introduce a new patch-based multi-resolution analysis of semi-regular mesh surfaces. This analysis brings patch-specific wavelet decomposition, quantization and encoding to the mesh compression process. Our underlying mesh partitioning relies on surface roughness (based on frequency magnitude variations), in order to produce patches, representative of semantic attributes of the object. With current compression methods based on wavelet decomposition, some regions of the mesh still have wavelet coefficients with a non negligible magnitude or polar angle (the angle with the normal vector), reflecting the high frequencies of the model. For each non-smooth region, our adaptive compression chain provides the possibility to choose the best prediction filter adjusted to its specificity. Our hierarchical analysis is based on a semi-regular mesh decomposition produced by second-generation wavelets. Apart from progressive compression, other types of applications can benefit from this adaptive decomposition, like error resilient compression, view-dependent reconstruction, indexation or watermarking. Selective refinement examples are given to illustrate the concept of ROI (Region Of Interest) decoding, which few people has considered, whereas it is possible with JPEG2000 for images.

This paper describes universal lossless coding strategies for compressing sources on countably infinite alphabets. Classes of memoryless sources defined by an envelope condition on the marginal distribution provide benchmarks for coding techniques originating from the theory of universal coding over finite alphabets. We prove general upper-bounds on minimax regret and lower-bounds on minimax redundancy for such source classes. The general upper bounds emphasize the role of the Normalized Maximum Likelihood codes with respect to minimax regret in the infinite alphabet context. Lower bounds are derived by tailoring sharp bounds on the redundancy of Krichevsky-Trofimov coders for sources over finite alphabets. Up to logarithmic (resp. constant) factors the bounds are matching for source classes defined by algebraically declining (resp. exponentially vanishing) envelopes. Effective and (almost) adaptive coding techniques are described for the collection of source classes defined by algebraically vanishing envelopes. Those results extend ourknowledge concerning universal coding to contexts where the key tools from parametric inference, 33 pages