Open-world Perception: A necessary path towards AGI

PhD Thesis Proposal Defence


Title: "Open-world Perception: A necessary path towards AGI"

by

Mr. Hao ZHANG


Abstract:

Open-world perception involves detecting and comprehending objects in
environments beyond the confines of training data. Traditional methods
primarily focus on identifying objects within predefined categories. In this
dissertation, we present a comprehensive approach to constructing a unified
architecture capable of recognizing and interpreting objects across various
contexts, responsive to user prompts. Initially, we delve into foundational
efforts aimed at enhancing the accuracy of object localization. Subsequently,
we explore the integration of open-vocabulary perception, leveraging language
as a conduit to broaden the object recognition vocabulary. Following this, we
describe strategies for tailoring perception to meet specific user needs,
guided by visual prompts. We conclude by highlighting the promising future of a
cohesive vision-language perception model, designed to adaptively detect and
interpret any object, fulfilling diverse user requirements.


Date:                   Monday, 19 February 2024

Time:                   10:00am - 12:00noon

Venue:                  Room 5501
                        Lifts 25/26

Committee Members:      Prof. Lionel Ni (Supervisor)
                        Prof. Harry Shum (Supervisor)
                        Dr. Dan Xu (Chairperson)
                        Dr. Qifeng Chen