Learning priors for neural scene representations
Abstract/Contents
- Abstract
- Neural 3D scene representations have recently emerged as a novel way to store information about 3D environments, their properties, and behavior, with learned features. Their key distinction from classical computer graphics representations is that the parameters of a representation can be obtained through optimization with the objective to be consistent with observations. As such, they have become a transformative tool for combining techniques from computer graphics and machine learning to represent 3D scenes. These representations have found use in applications ranging from robotics and remote sensing to cinematography and video editing. However, these representations are limited by the fact that they are only able to leverage captured information in a specific single scene. Specifically, only observations of a single individual scene can be used to create and improve the quality of the neural representation of this scene. This leads to undesirable properties, like slow creation of neural scene representations from observations or inability to generate realistic entirely new neural scene representations. In this thesis, Learning Priors for Neural Scene Representations, I propose novel methods for creating neural 3D scene representations which leverage information learned from data beyond a single individual scene. I look into using 3D, images, and pre-trained models as sources of information from which neural 3D scene representation priors can be learned. Moreover, I show how building standard computer graphics methodology into our neural scene representation architecture can ease the learning of priors from data which does not capture full 3D geometry.
Description
| Type of resource | text |
|---|---|
| Form | electronic resource; remote; computer; online resource |
| Extent | 1 online resource. |
| Place | California |
| Place | [Stanford, California] |
| Publisher | [Stanford University] |
| Copyright date | 2023; ©2023 |
| Publication date | 2023; 2023 |
| Issuance | monographic |
| Language | English |
Creators/Contributors
| Author | Bergman, Alexander William |
|---|---|
| Degree supervisor | Wetzstein, Gordon |
| Thesis advisor | Wetzstein, Gordon |
| Thesis advisor | Van Roy, Benjamin |
| Thesis advisor | Wu, Jiajun |
| Degree committee member | Van Roy, Benjamin |
| Degree committee member | Wu, Jiajun |
| Associated with | Stanford University, School of Engineering |
| Associated with | Stanford University, Department of Electrical Engineering |
Subjects
| Genre | Theses |
|---|---|
| Genre | Text |
Bibliographic information
| Statement of responsibility | Alexander William Bergman. |
|---|---|
| Note | Submitted to the Department of Electrical Engineering. |
| Thesis | Thesis Ph.D. Stanford University 2023. |
| Location | https://purl.stanford.edu/ks363mx0957 |
Access conditions
- Copyright
- © 2023 by Alexander William Bergman
- License
- This work is licensed under a Creative Commons Attribution 3.0 Unported license (CC BY).
Also listed in
Loading usage metrics...