Search

Your search keyword '"Luminari, Nicola"' showing total 12 results
Start Over Author "Luminari, Nicola"
12 results on '"Luminari, Nicola"'

1. DIAL: Deep Interactive and Active Learning for Semantic Segmentation in Remote Sensing

Author

Lenczner, Gaston, Chan-Hon-Tong, Adrien, Saux, Bertrand Le, Luminari, Nicola, and Besnerais, Guy Le

Subjects
Computer Science - Computer Vision and Pattern Recognition, Electrical Engineering and Systems Science - Image and Video Processing
Abstract

We propose in this article to build up a collaboration between a deep neural network and a human in the loop to swiftly obtain accurate segmentation maps of remote sensing images. In a nutshell, the agent iteratively interacts with the network to correct its initially flawed predictions. Concretely, these interactions are annotations representing the semantic labels. Our methodological contribution is twofold. First, we propose two interactive learning schemes to integrate user inputs into deep neural networks. The first one concatenates the annotations with the other network's inputs. The second one uses the annotations as a sparse ground-truth to retrain the network. Second, we propose an active learning strategy to guide the user towards the most relevant areas to annotate. To this purpose, we compare different state-of-the-art acquisition functions to evaluate the neural network uncertainty such as ConfidNet, entropy or ODIN. Through experiments on three remote sensing datasets, we show the effectiveness of the proposed methods. Notably, we show that active learning based on uncertainty estimation enables to quickly lead the user towards mistakes and that it is thus relevant to guide the user interventions.

Published
2022

2. Weakly-supervised continual learning for class-incremental segmentation

Author

Lenczner, Gaston, Chan-Hon-Tong, Adrien, Luminari, Nicola, and Saux, Bertrand Le

Subjects
Computer Science - Computer Vision and Pattern Recognition
Abstract

Transfer learning is a powerful way to adapt existing deep learning models to new emerging use-cases in remote sensing. Starting from a neural network already trained for semantic segmentation, we propose to modify its label space to swiftly adapt it to new classes under weak supervision. To alleviate the background shift and the catastrophic forgetting problems inherent to this form of continual learning, we compare different regularization terms and leverage a pseudo-label strategy. We experimentally show the relevance of our approach on three public remote sensing datasets. Code is open-source and released in this repository: https://github.com/alteia-ai/ICSS}{https://github.com/alteia-ai/ICSS.

Published
2022

3. Interactive Learning for Semantic Segmentation in Earth Observation

Author

Lenczner, Gaston, Chan-Hon-Tong, Adrien, Luminari, Nicola, Saux, Bertrand Le, and Besnerais, Guy Le

Subjects
Computer Science - Computer Vision and Pattern Recognition
Abstract

Dense pixel-wise classification maps output by deep neural networks are of extreme importance for scene understanding. However, these maps are often partially inaccurate due to a variety of possible factors. Therefore, we propose to interactively refine them within a framework named DISCA (Deep Image Segmentation with Continual Adaptation). It consists of continually adapting a neural network to a target image using an interactive learning process with sparse user annotations as ground-truth. We show through experiments on three datasets using synthesized annotations the benefits of the approach, reaching an IoU improvement up to 4.7% for ten sampled clicks. Finally, we exhibit that our approach can be particularly rewarding when it is faced to additional issues such as domain adaptation., Comment: 8 pages, 4 Figures, ECML-PKDD Workshop MACLEAN

Published
2020

4. DISIR: Deep Image Segmentation with Interactive Refinement

Author

Lenczner, Gaston, Saux, Bertrand Le, Luminari, Nicola, Tong, Adrien Chan Hon, and Besnerais, Guy Le

Subjects
Computer Science - Computer Vision and Pattern Recognition, Computer Science - Human-Computer Interaction
Abstract

This paper presents an interactive approach for multi-class segmentation of aerial images. Precisely, it is based on a deep neural network which exploits both RGB images and annotations. Starting from an initial output based on the image only, our network then interactively refines this segmentation map using a concatenation of the image and user annotations. Importantly, user annotations modify the inputs of the network - not its weights - enabling a fast and smooth process. Through experiments on two public aerial datasets, we show that user annotations are extremely rewarding: each click corrects roughly 5000 pixels. We analyze the impact of different aspects of our framework such as the representation of the annotations, the volume of training data or the network architecture. Code is available at https://github.com/delair-ai/DISIR., Comment: 8 pages, 12 figures. Published in the ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences

Published
2020
Full Text
View/download PDF

6. Autonomous Drones to Ensure Safety in Transport::Concept and Implementations

Author

Ebeid, Emad Samuel Malki, Jacobsen, Rune Hylsberg, Schenider-Kamp, Peter, Vom Bögel, Gerd, Guzzini, Paolo, Luminari, Nicola, Casarotti, Chiara, Mandirola, Martina, Taurino, Damiano, and Del Sole, Andrea

Abstract

Transport infrastructures require continuous and efficient inspection techniques to ensure their safety. Drones are one of the latest technologies that can be used to get a deep picture of the infrastructure. However, inspecting large infrastructures such as railways and bridges demand smart drone solutions that operate cooperatively and recharge autonomously while flying in a controlled framework. In this paper, we present the concept of the Horizon 2020 project Drones4Safety and its drone ecosystem for intelligent and autonomous inspection of linear infrastructures such as railways and bridges and discuss its integration with the U-Space.

Published
2022

9. Modeling and simulation of flows over and through fibrous porous media

Author

Luminari, Nicola, Institut National Polytechnique de Toulouse - INPT (FRANCE), and Institut National Polytechnique de Toulouse - Toulouse INP (FRANCE)

Subjects
Kriging metamodeling, Inertia, VANS, Porous media, CFD
Abstract

Any natural surface is in essence non-smooth, consisting of more or less regular roughness and/or mobile structures of different scales. From a fluid mechanics point of view, these natural surfaces offer better aerodynamic performances when they cover moving bodies, in terms of drag reduction, lift enhancement or control of boundary layer separation; this has been shown for boundary layer or wake flows around thick bodies. The numerical simulation of microscopic flows around "natural" surfaces is still out of reach today. Therefore, the goal of this thesis is to study the modeling of the apparent flow slip occurring on this kind of surfaces, modeled as a porous medium, applying Whitaker's volume averaging theory. This mathematical model makes it possible to capture details of the microstructure while preserving a satisfactory description of the physical phenomena which occur. The first chapter of this manuscript provides an overview of previous efforts to model these surfaces, detailing the most important results from the literature. The second chapter presents the mathematical derivation of the volume-averaged Navier-Stokes equations (VANS) in a porous medium. In the third chapter the flow stability at the interface between a free fluid and a porous medium, formed by a series of rigid cylinders, is studied. The presence of this porous layer is treated by including a drag term in the fluid equations. It is shown that the presence of this term reduces the rates of amplification of the Kelvin-Helmholtz instability over the whole range of wavenumbers, thus leading to an increase of the wavelength of the most amplified mode. In this same context, the difference between the isotropic model and a tensorial approach for the drag term has been evaluated, to determine the most consistent approach to study these flow instabilities. This has led to the conclusion that the model that uses the apparent permeability tensor is the most relevant one. In the following chapter, based on this last result, the apparent permeability tensor, based on over one hundred direct numerical simulations carried out over microscopic unit cells, has been identified for a three-dimensional porous medium consisting of rigid cylinders. In these configurations the tensor varies according to four parameters: the Reynolds number, the porosity and the direction of the average pressure gradient, defined by two Euler angles. This parameterization makes it possible to capture local three-dimensional effects. This database has been set up to create, based on a kriging-type approach, a behavioral metamodel for estimating all the components of the apparent permeability tensor. In the fifth chapter, simulations of the VANS equations are carried out on a macroscopic scale after the implementation of the metamodel, to get reasonable computing times. The validation of the macroscopic approach is performed on a closed cavity flow covered with a porous layer and a comparison with the results of a very accurate DNS, homogenized a posteriori, has shown a very good agreement and has demonstrated the relevance of the approach. The next step has been the study of the passive control of the separation of the flow past a hump which is placed on a porous wall, by the same macroscopic VANS approach. Finally, general conclusions and possible directions of research in the field are presented in the last chapter.

Published
2018

11. Sensitivity and stability analysis of a rigid canopy flow

Author

Luminari, Nicola, Airiau, Christophe, Association Française de Mécanique, Service irevues, irevues, Centre National de la Recherche Scientifique - CNRS (FRANCE), Institut National Polytechnique de Toulouse - INPT (FRANCE), Université Toulouse III - Paul Sabatier - UT3 (FRANCE), and Institut National Polytechnique de Toulouse - Toulouse INP (FRANCE)

Subjects
canopy, Sensitivity, adjoint, Mécanique des fluides, Canopy, [PHYS.MECA]Physics [physics]/Mechanics [physics], [PHYS.MECA] Physics [physics]/Mechanics [physics], stability, sensitivity, Stability
Abstract

Colloque avec actes et comité de lecture. Internationale.; International audience; Le projet a pour objectif d'étudier un écoulement autour d'une paroi recouverte de filaments flexibles en vue d'obtenir une réduction de traînée. On regarde d'abord un modèle de canopée rigide qui comprend un grand nombre de paramètres. Une étude de sensibilité est menée sur la base d'équations adjointes pour déterminer les paramètres les plus influents sur la forme du profil de vitesse et sur des quantités intégrales. On étudie ensuite la stabilité locale temporelle de l'écoulement comprenant la canopée et la partie externe. Enfin on détermine la sensibilité des valeurs propres les plus instables associée à une modification de l'écoulement moyen qui provient éventuellement d'une variation des paramètres. Cette approche est basée sur les équations adjointes de stabilité.

Published
2015

Catalog

Books, media, physical & digital resources
See catalog results