Scientific Publications of the University of Toulouse II Le Mirail
Not a member yet
92205 research outputs found
Sort by
Power Cost Comparison of Neural-Network Compression Methods for Satellite Imagery
No full textInternational audienc
Deep Priors for Satellite Image Restoration With Accurate Uncertainties
No full textInternational audienceSatellite optical images, upon their on-ground receipt, offer a distorted view of the observed scene. Their restoration, including denoising, deblurring, and sometimes superresolution, is required before their exploitation. Moreover, quantifying the uncertainties related to this restoration helps to reduce the risks of misinterpreting the image content. Deep learning methods are now state-of-the-art for satellite image restoration. Among them, direct inversion methods train a specific network for each sensor, and generally provide a point estimation of the restored image without the associated uncertainties. Alternatively, deep regularization (DR) methods learn a deep prior on target images before plugging it, as the regularization term, into a modelbased optimization scheme. This allows for restoring images from several sensors with a single network and possibly for estimating associated uncertainties. In this paper, we introduce VBLE-xz, a DR method that solves the inverse problem in the latent space of a variational compressive autoencoder (CAE). We adapt the regularization strength by modulating the bitrate of the trained CAE with a training-free approach. Then, VBLE-xz estimates relevant uncertainties jointly in the latent and in the image spaces by sampling an explicit posterior estimated within variational inference. This enables fast posterior sampling, unlike state-of-the-art DR methods that use Markov chains or diffusion-based approaches. We conduct a comprehensive set of experiments on very high-resolution simulated and real Pléiades images, asserting the performance, robustness and scalability of the proposed method. They demonstrate that VBLE-xz represents a compelling alternative to direct inversion methods when uncertainty quantification is required. The code associated to this paper is available in https://github.com/MaudBqrd/VBLExz
Les modes d’engagement des dirigeants bénévoles en association sportive : entre conversion de la carrière sportive, stratégies familiales et cooptation
No full textInternational audienceLa recherche vise à appréhender les modes d’engagement des dirigeants bénévoles en association sportive. À partir de la sociologie des carrières, il s’agit de comprendre comment naissent, se prorogent, se réorientent ou s’interrompent les carrières de ces acteurs essentiels et indispensables au monde sportif amateur. Suite au recueil de données qualitatives réalisé au sein d’un petit club de rugby, nous avons mis en évidence deux manières de s’engager dans la carrière de dirigeant bénévole sportif, selon les positions occupées, les motivations et valeurs transmises, et les représentations de leurs rôles. Ont également été considérés les effets des contextes familiaux et professionnels sur l’engagement bénévole
Paradigms in Morphology
No full textDuring the 20th century, morphology has long been described in terms of rules and processes. In recent decades, the focus of inflectional morphology has shifted from the structure of units to the structures they form together. These structures are paradigms containing lexemes that vary in the same way. With the aim of unifying inflectional and derivational morphology, several proposals have been made to transfer inflectional paradigms to derivation, taking advantage of the many similarities between the two subfields of morphology
Pour une approche intégrative de la santé: Perspectives croisées en SHS. Actes du 1er colloque international ICARES
No full textInternational audienceLes sciences humaines et sociales (SHS) jouent un rôle essentiel dans la compréhension et l’amélioration de la santé globale. Elles permettent d’analyser les déterminants sociaux, culturels, économiques et politiques qui influencent les inégalités de santé à l’échelle mondiale. Elles éclairent les pratiques de soin, les politiques publiques, les perceptions du risque ou encore l’adhésion aux campagnes de prévention. Face à des crises sanitaires globales, elles offrent des outils d’analyse critique, de médiation et d’accompagnement des populations, en plaçant l’expérience humaine au cœur des réponses. Dans cet ouvrage, les contributions visent à promouvoir une interdisciplinarité active, inscrite en complémentarité avec les disciplines médicales. Elles favorisent un dialogue ouvert et critique avec les communautés ainsi que des interventions pour réduire les inégalités d’accès à la santé. Le premier colloque international d’ICARES, s’est attelé à démontrer que les SHS participent à construire une approche plus intégrative, équitable et complexe de la santé. ICARES est l’Institut pour la Compréhension, l’Anticipation et l’inteRvention en Environnement et Santé globale porté par l’université de Montpellier Paul-Valéry
Results of the Ontology Alignment Evaluation Initiative 2025
No full textInternational audienceThe Ontology Alignment Evaluation Initiative (OAEI) aims at comparing ontology matching systems on precisely defined test cases. These test cases can be based on ontologies of different levels of complexity and use different evaluation modalities. The OAEI 2025 campaign offered 12 tracks and was attended by 20 participants. This paper is an overall presentation of that campaign
La diversité du sentiment de reconnaissance chez les conseillers/référents RSA
No full textInternational audienc
Local Weather Types (LWTs) associated with Urban Heat Islands (UHIs) and hot days in the Grenoble area, France
No full textInternational audienceUrban heat extremes are intensified by both global warming and the urban heat island (UHI) effect, particularly in summer. This study investigates the meteorological conditions driving extreme summer temperatures in Grenoble (French Prealps), a city with complex topography. The objectives are threefold: 1/ Assess the performance of the local weather types (LWTs) classification method in this context; 2/ Adapt and optimizing the method for topographically and climatically heterogeneous environments and 3/ Identify LWTs associated with extreme heat events—heatwaves, tropical nights and UHIs—for operational used by local authorities. Daily meteorological data (precipitation, temperature range, wind speed, wind direction and specific humidity), which are utilized for classification, are sourced from: (i) ERA5 reanalysis at the nearest the rural grid point (1960–2001 and 2002–2022), compensating for the lack of long-term representative data for the area; and (ii) a dense urban observation network (2019–2022). In this study, in order to better represent the complexity of meteorological conditions due to the incised topography, the original LWT method is modified. Methodological adjustments include discretizing precipitation into three categories and excluding wind direction, shown to have negligible influence due to topographic constraints. Results demonstrate that the modified LWT method effectively captures local meteorological variability and is transferable to similar contexts. LWTs linked to extreme heat events in Grenoble are characterized by high diurnal temperature ranges, dry conditions, elevated specific humidity, and low wind speeds