« Zurück

Lettry Louis

Professeur-e HES Assistant-e

Hauptkompetenzen

Professeur-e HES Assistant-e

Büro: ENP.23.N317

HES-SO Valais-Wallis - Haute Ecole d'Ingénierie
Rue de l'Industrie 23, 1950 Sion, CH

Bereich
Technique et IT

Hauptstudiengang
Informatique et systèmes de communication

BSC HES-SO en Informatique et systèmes de communication - HES-SO Valais-Wallis - Haute Ecole d'Ingénierie

Traitement d'image et vision
Algorithms and Data Structures
Cryptographie

Abgeschlossen

WhiteNightSkEyes - Chèque Inno Contrat no 64871.1 INNO-ICT

AGP

Rolle: Hauptgesuchsteller/in

Financement: Innosuisse - cheques

Description du projet : Sean Lally Architecture Sàrl is an architecture bureau created by Professor Sean Lally and versed in virtual reality. It aims to present architecture and its creations in a dynamic and novel fashion to be enjoyed by all. Through a mix of augmented reality in previous projects (The Long Now) and virtual reality (Shaped Touches), their aim has been to propose different ways of perceiving and interacting between architects, their creations and the general public. Currently, the bureau focuses on creating a set of tools to produce immersive virtual worlds to be experienced. The end goal is the creation of an online virtual talk show with live host, guests and audience, all participating from home in these designed spaces. This would create another level of interactivity between architects, clients and guests that would be of interest to many others beyond this specific project. Parts of this project are inherently immersive and straightforward to produce, for example bringing a building into a virtual world, others are much more complex to achieve and outside the expertise of the bureau. The part that this project concerns relates to bringing guests in an immersive manner into the virtual world. Current options which include a streaming webcam video plane presents very obvious limitations on experience and immersion. The research would focus on evaluating the different methods that the research collaborator could propose and suggest to bring 3D virtual humans in the talk show.

Forschungsteam innerhalb von HES-SO: Bertaud Adrien , Lettry Louis

Partenaires académiques: VS - Institut Systèmes industriels

Durée du projet: 22.12.2022 - 22.12.2023

Montant global du projet: 13'928 CHF

Statut: Abgeschlossen

Calligranalytics - No 63755.1 INNO-ICT

AGP

Rolle: Hauptgesuchsteller/in

Financement: Innosuisse

Description du projet : Le processus actuel de numérisation de caractères demande beaucoup de temps ainsi que l'expertise d'un opérateur bien formé. Il gagnerait donc à être automatisé. Cela permettra non seulement de réduire les coûts mais aussi d'en accroitre l'accessibilité des résultats et leur précision. La question principale de cette étude consiste à évaluer et implémenter différentes méthodes d'automatisations pour numériser des caractères manuscrits à boucles (tels que les "o", "d", ...) en des squelettes analysables par les outils déjà développés par OrphAnalytics. La variété des styles d'écritures et des conditions de rédaction rend ce problème particulièrement complexe. De plus une évaluation des potentielles propriétés extractibles grâce à ces méthodes complètera cette étude.

Forschungsteam innerhalb von HES-SO: Lettry Louis

Partenaires académiques: VS - Institut Systèmes industriels

Durée du projet: 27.10.2022 - 27.06.2023

Montant global du projet: 15'000 CHF

Statut: Abgeschlossen

2018

Unsupervised Deep Single‐Image Intrinsic Decomposition using Illumination‐Varying Image Sequences

Wissenschaftlicher Artikel

Lettry Louis, Vanhoey Kenneth, Luc van Gool

Pacific Graphics (PG), 2018

Link zur Publikation

Zusammenfassung:

Machine learning based Single Image Intrinsic Decomposition (SIID) methods decompose a captured scene into its albedo and shading images by using the knowledge of a large set of known and realistic ground truth decompositions. Collecting and annotating such a dataset is an approach that cannot scale to sufficient variety and realism. We free ourselves from this limitation by training on unannotated images. Our method leverages the observation that two images of the same scene but with different lighting provide useful information on their intrinsic properties: by definition, albedo is invariant to lighting conditions, and cross-combining the estimated albedo of a first image with the estimated shading of a second one should lead back to the second one's input image. We transcribe this relationship into a siamese training scheme for a deep convolutional neural network that decomposes a single image into albedo and shading. The siamese setting allows us to introduce a new loss function including such cross-combinations, and to train solely on (time-lapse) images, discarding the need for any ground truth annotations. As a result, our method has the good properties of i) taking advantage of the time-varying information of image sequences in the (pre-computed) training step, ii) not requiring ground truth data to train on, and iii) being able to decompose single images of unseen scenes at runtime. To demonstrate and evaluate our work, we additionally propose a new rendered dataset containing illumination-varying scenes and a set of quantitative metrics to evaluate SIID algorithms. Despite its unsupervised nature, our results compete with state of the art methods, including supervised and non data-driven methods.

DARN: A Deep Adversarial Residual Network for Intrinsic Image Decomposition

Wissenschaftlicher Artikel

Lettry Louis, Kenneth Vanhoey, Luc van Gool

Winter Conference on Applications of Computer Vision (WACV), 2018

Link zur Publikation

Zusammenfassung:

We present a new deep supervised learning method for intrinsic decomposition of a single image into its albedo and shading components. Our contributions are based on a new fully convolutional neural network that estimates absolute albedo and shading jointly. Our solution relies on a single end-to-end deep sequence of residual blocks and a perceptually-motivated metric formed by two adversarially trained discriminators. As opposed to classical intrinsic image decomposition work, it is fully data-driven, hence does not require any physical priors like shading smoothness or albedo sparsity, nor does it rely on geometric information such as depth. Compared to recent deep learning techniques, we simplify the architecture, making it easier to build and train, and constrain it to generate a valid and reversible decomposition. We rediscuss and augment the set of quantitative metrics so as to account for the more challenging recovery of non scale-invariant quantities. We train and demonstrate our architecture on the publicly available MPI Sintel dataset and its intrinsic image decomposition, show attenuated overfitting issues and discuss generalizability to other data. Results show that our work outperforms the state of the art deep algorithms both on the qualitative and quantitative aspect.

Wildtrack: A multicamera HD dataset for dense unscripted pedestrian detection

Wissenschaftlicher Artikel

Tatjana Chavdarova, Pierre Baqué, Stéphane Bouquet, Andrii Maksai, Jose Cijo, Timur Bagautdinov, Lettry Louis, Pascal Fua, Luc van Gool, François Fleuret

Conference on Computer Vision and Pattern Recognition (CVPR), 2018

Link zur Publikation

Zusammenfassung:

People detection methods are highly sensitive to occlusions between pedestrians, which are extremely frequent in many situations where cameras have to be mounted at a limited height. The reduction of camera prices allows for the generalization of static multi-camera set-ups. Using joint visual information from multiple synchronized cameras gives the opportunity to improve detection performance. In this paper, we present a new large-scale and high-resolution dataset. It has been captured with seven static cameras in a public open area, and unscripted dense groups of pedestrians standing and walking. Together with the camera frames, we provide an accurate joint (extrinsic and intrinsic) calibration, as well as 7 series of 400 annotated frames for detection at a rate of 2 frames per second. This results in over 40 000 bounding boxes delimiting every person present in the area of interest, for a total of more than 300 individuals. We provide a series of benchmark results using baseline algorithms published over the recent months for multi-view detection with deep neural networks, and trajectory estimation using a non-Markovian model.

2017

Repeated pattern detection using CNN activations

Wissenschaftlicher Artikel

Lettry Louis, Michal Perdoch, Kenneth Vanhoey, Luc van Gool

Winter Application of Computer Vision (WACV), 2017

Link zur Publikation

Zusammenfassung:

We propose a new approach for detecting repeated patterns on a grid in a single image. To do so, we detect repetitions in the space of pre-trained deep CNN filter response at all layer levels. These encode features at several conceptual levels (from low-level patches to high-level semantics) as well as scales (from local to global). As a result, our repeated pattern detector is robust to challenging cases where repeated tiles show strong variation in visual appearance due to occlusions, lighting or background clutter. Our method contrasts with previous approaches that rely on keypoint extraction, description and clustering or on patch correlation. These generally only detect low-level feature clusters that do not handle variations in visual appearance of the patterns very well. Our method is simpler, yet incorporates high level features implicitly. As such, we can demon strate detections of repetitions with strong appearance variations, organized on a nearly-regular axis-aligned grid Results show robustness and consistency throughout a varie database of more than 150 images.

2016

Markov chain monte carlo cascade for camera network calibration based on unconstrained pedestrian tracklets

Wissenschaftlicher Artikel

Lettry Louis, Ralf Dragon, Luc van Gool

Asian Conference on Computer Vision, 2016

Link zur Publikation

Zusammenfassung:

The presented work aims at tackling the problem of externally calibrating a network of cameras by observing a dynamic scene composed of pedestrians. It relies on the single assumption that human beings walk aligned with the gravity vector. Usual techniques to solve this problem involve using more assumptions such as a planar ground or assumptions about pedestrians’ motion. In this work, we drop all these assumptions and design a probabilistic layered algorithm that deals with noisy outlier-dominated hypotheses to recover the actual structure of the network. We demonstrate our process on two known public datasets and exhibit results to underline the effectiveness of our simple but adaptable approach to this general problem.

Errungenschaften

PEOPLE@HES-SO
Verzeichnis der Mitarbeitenden und Kompetenzen

Lettry Louis

Professeur-e HES Assistant-e

Hauptkompetenzen

Deep Learning

Computer vision

Professeur-e HES Assistant-e

PEOPLE@HES-SO Verzeichnis der Mitarbeitenden und Kompetenzen

Lettry Louis

Professeur-e HES Assistant-e

Hauptkompetenzen

Deep Learning

Computer vision

Professeur-e HES Assistant-e

PEOPLE@HES-SO
Verzeichnis der Mitarbeitenden und Kompetenzen