Canada Spends Logo

Explorez notre analyse du budget 2026

Grant: Discovery Grants Program - Individual

University of Waterloo

$36,000.00

Weakly Supervised and Unsupervised Semantic Segmentation with Regularized Loss

EF 2024-25

Résumé
The proposed research lies in the area of computer vision called semantic segmentation. In semantic segmentation, given an image, the goal is to assign every pixel a label indicating the object class. Semantic segmentation has multiple applications such as autonomous driving, medical imaging and diagnostics, photovideo editing, etc. Convolutional Neural Networks (CNNs) lead to a breakthrough in semantic segmentation in the fully supervised setting. The fully supervised setting requires datasets with thousands of images where each pixel is labeled with its object class. Constructing such datasets is extremely costly in terms of manual labor, and is a serious bottleneck in semantic segmentation. The goal of proposed research is to develop semantic segmentation methods with imprecise or only partially labeled data, which is called weakly supervised segmentation and also in segmentation without any user annotation, which is called unsupervised segmentation. This will be achieved through the pursuit of the following objectives: (I) Image Level Multi-class Semantic Segmentation with Regularized Loss; (II) Improved CRF models; (III) Improved Minimization of Regularized Loss; (IV) Unsupervised Semantic Segmentation. The underlying theme of my approach is using CRF models for developing loss functions suitable for the weakly supervised and unsupervised settings. This builds on my prior expertise in Conditional Random Fields (CRFs), which are classical, statistically justified models. Since CRFs are successful at modelling a priori likely properties of objects, they are particularly suitable for object discovery in weakly supervised or unsupervised segmentation. The methods in this proposal will improve the state of the art in semantic segmentation, ease the requirements of manual labeling and will lead to faster deployment in practical computer vision applications where semantic segmentation is required. There will be 7 undergraduate, 7 master, 2 doctoral students as well as 1 Postdoctoral fellow trained through this proposal. The trainees will gain expertise in computer vision, discrete optimization, and deep learning. These skills that are highly in demand in academia and industry, with the recent advances in the field of Artificial Intelligence, which already resulted in many practical applications brought to everyday lives, and is well posed to bring more applications in the future.
Programme
Discovery Grants Program - Individual
Emplacement
Ontario, Canada
Type de bénéficiaire
Academia
Type
Grant
Source
NSERC's Awards Data

Tel que publié

Application id
RGPIN-2023-03776
Name nom
Veksler, Olga
Department département
Computer Science, David R. Cheriton School of - Computer Science, David R. Cheriton School of
Organization id
33
Institution établissement
University of Waterloo
Province en
Ontario
Province fr
Ontario
Country en
CANADA
Country fr
CANADA
Fiscal year exercice financier
2024
Competition year année de concours
2023
Award amount
36000
Program id
RGPIN
Program name en
Discovery Grants Program - Individual
Program name fr
Programme de subventions à la découverte - individuelles
Group en
DISCOVERY RESEARCH
Group fr
RECHERCHE AXÉE SUR LA DÉCOUVERTE
Committee code
1507
Committee name en
Computer Science
Committee name fr
Sciences informatiques
Area of application code
1207
Area of application group en
Advancement of knowledge
Area of application group fr
Avancement des connaissances
Area of application en
Information, computer and communication technologies
Area of application fr
Technologie de l'information, des ordinateurs et des réseaux de communications
Research subject code
2700
Research subject group en
Information technology
Research subject group fr
Technologie de l'information
Research subject en
Information technology
Research subject fr
Technologie de l'information
Field of research list names en
Not available
Field of research list names fr
Non disponible
Socioeconomic objective list names en
Not available
Socioeconomic objective list names fr
Non disponible
Application title
Weakly Supervised and Unsupervised Semantic Segmentation with Regularized Loss
Keywords
computer vision; machine learning; semantic segmentation; conditional random fields; deep learning; weak supervision
Application summary
The proposed research lies in the area of computer vision called semantic segmentation. In semantic segmentation, given an image, the goal is to assign every pixel a label indicating the object class. Semantic segmentation has multiple applications such as autonomous driving, medical imaging and diagnostics, photovideo editing, etc. Convolutional Neural Networks (CNNs) lead to a breakthrough in semantic segmentation in the fully supervised setting. The fully supervised setting requires datasets with thousands of images where each pixel is labeled with its object class. Constructing such datasets is extremely costly in terms of manual labor, and is a serious bottleneck in semantic segmentation. The goal of proposed research is to develop semantic segmentation methods with imprecise or only partially labeled data, which is called weakly supervised segmentation and also in segmentation without any user annotation, which is called unsupervised segmentation. This will be achieved through the pursuit of the following objectives: (I) Image Level Multi-class Semantic Segmentation with Regularized Loss; (II) Improved CRF models; (III) Improved Minimization of Regularized Loss; (IV) Unsupervised Semantic Segmentation. The underlying theme of my approach is using CRF models for developing loss functions suitable for the weakly supervised and unsupervised settings. This builds on my prior expertise in Conditional Random Fields (CRFs), which are classical, statistically justified models. Since CRFs are successful at modelling a priori likely properties of objects, they are particularly suitable for object discovery in weakly supervised or unsupervised segmentation. The methods in this proposal will improve the state of the art in semantic segmentation, ease the requirements of manual labeling and will lead to faster deployment in practical computer vision applications where semantic segmentation is required. There will be 7 undergraduate, 7 master, 2 doctoral students as well as 1 Postdoctoral fellow trained through this proposal. The trainees will gain expertise in computer vision, discrete optimization, and deep learning. These skills that are highly in demand in academia and industry, with the recent advances in the field of Artificial Intelligence, which already resulted in many practical applications brought to everyday lives, and is well posed to bring more applications in the future.