Facilitating deep learning in computer vision by adding hyperspectral and polarity information
We conjecture that progress in artificial perception in outdoor environments is currently impeded by focusing on RGB images. In this project, we seek to expand the sensory information available to train such visual perception systems, by relying on two newly available sensors: a multipolarized camera and a snapshot hyperspectral camera (50 to 204 spectral bands). Our working hypothesis is that the availability of this spectral and polarity information during training on RGB images of the exact same scene will facilitate learning. This project thus aims at making significant and fundamental contributions to the artificial perception problem, by demonstrating the importance of sensing in this complex equation.
As such, part of the project will entail building a data capture rig with the above-mentioned sensors, coupled with a high-resolution RGB camera. Then, you would be responsible for gathering and curating this unique dataset. In a second phase, you will seek how to exploit this rich multimodal information to facilitate deep learning. In particular, self-supervised learning techniques will be explored, unlocking new approaches to reduce the need for labelled data, a pressing issue in Deep Learning. Architectural development to handle hyperspectral information will also be a key component of the project.
- Artificial intelligence
- Deep Learning
- Self-supervised algorithms
- Computer vision
- Hyperspectral and RGB images
The Northern Robotics Laboratory (Norlab) has over twenty interns, students and research professionals seeking to push ahead the state of the art in mobile robotics and artificial perception in outdoor and challenging environments. Members are regularly publishing in highly competitive venues, such as ICRA, IROS, WACV, etc. The Norlab is well-equipped, with numerous all-terrain autonomous vehicles (Warthog, Husky, Superdroid), many 3D sensors, specialized cameras (two hyperspectral, two multi-polarized and one high-resolution 20 MPix) as well as a solid foundation of computing equipment, in terms of GPUs (NVIDIA 1xA100, 4xA6000, and 3x3090).
Financial Aid Available by Program of Study
Doctorate in Computer Science
Financial Aid Available*
Financial Aid Related to Research Project
$21000 per year for 4 years.
Program-Specific Financial Aid
Graduate Studies Awards
|7 x $500
|Doctoral exam or seminar presentation
|Doctoral exam, seminar or conference
|First submission of dissertation
*A $500 bonus will be added to the graduation scholarship if the initial submission is made before the semester preceding the final qualifying semester.
Université Laval: Student Financial Aid
Supplemental Tuition Fee Exemption Scholarship Program: Entitles international students to pay Canadian student tuition fees, for overall savings of around $40,000. If you are benefiting of this scholarship program, the doctoral admission scholarship will go toward paying your supplemental tuition fees. You will not receive the $2,000 scholarship amount directly; rather it will be applied over your first two semesters to the supplemental portion of your tuition fees.
* Amounts shown represent maximum financial aid available. Certain conditions apply. Subject to change without prior notice. For further information, contact sponsoring organizations directly.
- Computer Software
- Electrical Engineering
- Physics Engineering
- Computer Engineering
- Software Engineering
- Artificial intelligence
Requirements and Conditions
- Good knowledge in Python
- Excellent thinking abilities
- Natural curiosity
- Motivation and self-driven
- Able to write high-quality documents
- Curriculum vitæ
- Student transcript
Find Out More
Département d'informatique et de génie logiciel