Text-to-Video synthesis for 3D reconstruction
Project Description
Students will learn how to perform text-to-video synthesis or generate consecutive frames with consistent camera poses for downstream 3D reconstruction and scene understanding. We focus on generating multiple consistent and consecutive frames. Based on generated videos, we perform 3D modeling, which requires a strong ability to generate outputs with 3D geometry consistency.
Supervisor
YEUNG, Sai Kit
Quota
5
Course type
UROP1000
UROP1100
UROP2100
UROP3100
UROP3200
UROP4100
Applicant's Roles
Students will learn how to use the recent powerful text-to-video techniques for customized generation with consistent geometric consistency. Students will also learn some basic usages of 3D reconstruction software and learn the basics of 3D scene understanding.
Applicant's Learning Objectives
Get the ability to use text-to-video techniques for both 2D and 3D content generation.
Complexity of the project
Challenging