Retour à la programmation
Lundi 4 avril
13:00 à 14:20
Activité Rendez-vous IA Québec (RVIAQ)
  • Mode rediffusion
  • Langue
  • Lieu Scène Google Cloud

Relève en intelligence et données - Axe environnement physique

13h - Présentation d'Ali Ahmadi (en anglais)

Smart Sidewalk : Using Deep Learning for Mobility Assessment of Wheelchair Users in Winter Conditions from Multisensor Data

This talk aims at briefly presenting the results of a pilot project called “smart sidewalk” and carried out in partnership with a Quebec-based company called Kalitec providing the research team with access to its innovative set of self-sufficient terminals composed of different sensors (air and surface condition sensors, camera with motion detectors, and communication equipment, etc.). These sensors were installed in an intersection in front of the rehabilitation institute in Quebec City (IRDPQ) to analyze the sidewalk accessibility and the mobility of wheelchair users in winter conditions. The system allowed acquiring huge amount of data in the intersection for 6 months (1 image per 15 minutes together with data on air and surface temperature and humidity conditions). To detect and analyze the mobility of wheelchair users, we applied a multilayer deep learning algorithm called YOLOv3 (You Look Only Once version 3). During our talk, we will give some insight into the obtained results and will present some future research and development orientations.

13h20 - Présentation d'Aymane Dahbi (en français)

Vers une nouvelle approche d’extraction des réseaux hydrographiques à partir des points LiDAR massifs au sol

L'objectif de notre recherche consiste à développer une nouvelle approche d'extraction des réseaux hydrographiques directement à partir des points LiDAR massif au sol. Cette approche repose sur la théorie de Morse pour le calcul du réseau de surface caractérisant le squelette des réseaux hydrographiques. Elle prendra également en compte le défi lié au traitement des données LiDAR massives à travers l’implémentation d’une solution de parallélisme par conteneurs afin de garantir l'extensibilité de l'approche proposée.

13h40 - Présentation de Pauline Perbet (en français)

Utiliser l’apprentissage profond pour relever le défi de la classification du type de perturbation forestière par télédétection

La classification du type de perturbation forestière par télédétection reste un défi, en particulier en ce qui concerne les perturbations partielles (coupe ou feux partiels) ou progressives (épidémie). Notre projet évalue la capacité des modèles d’apprentissages profonds à tirer parti des caractéristiques spectrales et temporelles de la collection d’images Landsat pour améliorer la qualité des cartographies de perturbation au Québec.

14h - Présentation de Mohammad Reza Karimi Dastjerdi (en anglais)

Guided Co-Modulated GAN for 360° Field of View Extrapolation

We propose a method to extrapolate a 360° field of view from a single image that allows for user-controlled synthesis of the out-painted content. To do so, we propose improvements to an existing GAN-based in-painting architecture for out-painting panoramic image representation. Our method obtains state-of-the-art results and outperforms previous methods on standard image quality metrics. To allow controlled synthesis of out-painting, we introduce a novel guided co-modulation framework, which drives the image generation process with a common pre trained discriminative model. Doing so maintains the high visual quality of generated panoramas while enabling user-controlled semantic content in the extrapolated field of view. We demonstrate the state-of-the-art results of our method on field of view extrapolation both qualitatively and quantitatively, providing thorough analysis of our novel editing capabilities. Finally, we demonstrate that our approach benefits the photorealistic virtual insertion of highly glossy objects in photographs.

Ali Ahmadi | Étudiant au doctorat en sciences géomatiques | Université Laval

Ali Ahmadi is a PhD student in the Department of Geomatics Sciences at Université Laval. He is also a member of CRDIG and CIRRIS. His research interests include geographic information sciences, human environment interactions and inclusive mobility. In his PhD project, he works on the design and development of novel algorithms from artificial intelligence and multisensory data fusion to assess the accessibility of urban places for the benefit of people with motor disabilities difficulties

Pauline Perbet | Etudiante | Université Laval

Pauline Perbet est étudiante au doctorat depuis deux ans. Avant cela elle a réalisé une maitrise en recherche aussi à l’Université Laval, sous la direction de Martin Béland. Dans une autre vie, elle a eu un master en science environnementale en France, puis a travaillé en tant quegGéomaticienne pendant plusieurs années dans un parc national en Guyane Francaise.

Mohammad Reza Karimi Dastjerdi | Étudiant au doctorat en génie électrique | Université Laval

Mohammad Reza Karimi Dastjerdi is a 3rd-year Ph.D. candidate in electrical engineering at Université Laval, supervised by Professor Jean-François Lalonde. He is a member of Institut intelligence et données (IID). Mohammad's main research interests are lighting estimation, novel view synthesis, and scene understanding. During his Ph.D. at Université Laval, Mohammad had the opportunity to work with Adobe as a research intern, in which he was working on the Image Match feature of Adobe Dimension. Before joining Université Laval, Mohammad obtained an M.Sc. in culture technology in 2019 at the Korea advanced institute of Science and Technology (KAIST), Deajeon, South Korea. He worked on generative adversarial networks for video generation under the supervision of Junyong Noh. Mohammad received a B.Sc. in software engineering in 2015 at the K.N.Toosi University of Technology, Tehran, Iran.

Logo de CanExplore Logo de Norda Stelo Solutions Logo de Rendez-vous IA Québec