Back to search

Cornell University

Neural Representation for Camera Models and Dynamic Scenes

Abstract

dc:description.abstract

The advent of digital and smartphone cameras has revolutionized photography by making it accessible to almost everyone. Computational photography offers exciting opportunities to democratize creative photography, allowing beginners to create amazing visual content effortlessly. This is achieved by leveraging computer vision algorithms and machine learning to enable users to create high-quality visual content effortlessly. In my thesis dissertation, I will discuss three novel approaches we proposed for representing camera models and 3D scenes using deep neural networks. The first part of this dissertation focuses on the development of a neural lens model, which simplifies the process of pre-capture camera calibration and refines camera parameters during 3D reconstruction. The proposed method accurately models the effects of the camera lens distortion using a neural network, which can be easily integrated into downstream neural rendering systems. My second work focuses on camera orientation estimation from single images by incorporating 3D geometry. Our proposed system can be trained end-to-end with camera poses and intermediate representations of surface geometry, outperforming a black-box regression from image to camera parameters. Third, I will present a compelling application of my research on creating 3D videography from input videos casually captured by cellphones. By using such algorithms, we can create immersive experiences that were previously only possible with specialized equipment and expertise. Lastly, I will discuss my current and upcoming research, which seek to expand the possibilities of artistic expression and content creation. I will highlight some exciting directions in the future, through enhancing scene rendering quality, implementing novel features like AR/VR and visual effects, and ultimately providing users with a more effortless and creative photography experience.

Degree

thesis:*
Name thesis:degree_name
Ph. D., Computer Science
Level thesis:degree_level
Doctor of Philosophy
Discipline thesis:degree_discipline
Computer Science
Grantor
Cornell University
Year dc:date.issued
2023

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Xian, Wenqi
Chair dc:contributor.chair
  • Snavely, Keith
Committee members dc:contributor.committeemember
  • Belongie, Serge
  • Lee, Clarence

Subjects

dc:subject × 3

Rights

dc:rights
Statement dc:rights
  • Attribution 4.0 International
Language dc:language.iso
en

Identifiers

dc:identifier.*
Dc Identifier Bibid
16176501
Dc Identifier Other
ProQuest Submission ID: 13651
ProQuest Publication ID: 30485933

Chain of custody

source
Harvested from
Cornell University
Base URL
ecommons.cornell.edu/server/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
citation

Xian, Wenqi. Neural Representation for Camera Models and Dynamic Scenes. Doctor of Philosophy thesis, Cornell University, 2023. https://hdl.handle.net/1813/114177