Interactive Annotations of Large-Scale Image Data for Training of Recognition Models

The main objective of this project is to investigate, develop and evaluate state-of-the-art color image/video analysis and machine-learning algorithms, which are suitable for accurate modeling and recognition from large-scale image datasets that are weakly labeled. In particular, we will focus on investigating and developing interactive (partially-supervised) algorithms for annotating massive sets of color images, while minimizing the user efforts. The ultimate goal is to build stat-of-the-art color image classification models, a computer vision application area of high interest to QuindiTech. The specific objectives are: (1) Interactive object delineations in massive image/video data sets; and (2) Active categorization of images at a web-scale level. Learning weakly supervised image recognition models typically leads to complex and ill-posed optimization problems. This project will leverage some limited and targeted interactions with humans, as needed, to set optimization constraints and to drive advanced learning methods. TO BE CONT’D

Faculty Supervisor:

Ismail Ben Ayed;Éric Granger

Student:

Partner:

QuindiTech

Discipline:

Engineering

Sector:

Professional, scientific and technical services

University:

École de technologie supérieure

Program: