Modeling virtual environments 3D assets from visual inputs
Abstract/Contents
- Abstract
- Modeling the physical world is crucial for many applications that significantly impact society, includ- ing robotics, autonomous vehicles, virtual and augmented reality (AR/VR). Research works com- monly leverage 3D virtual environments as a proxy for real environments to perform experiments. By using virtual environments, experiments like robotic simulations, navigation, and interactions can be performed safely and at a lower cost compared to real-world environments. 3D geometric and material assets are the building blocks of virtual environments. However the process of creating 3D assets from scratch is mostly manual, time-consuming, and requires significant expertise. This challenge motivates the development of automated methods for 3D assets generation. We introduce a Computer Vision framework for generating 3D assets from visual inputs that include images and 3D point clouds from real scenes. Visual inputs from a scene are parsed into individual objects, structures, and materials which can be processed into 3D assets. This framework involves four key tasks: 3D semantic segmentation, shape completion, shape modeling, and material prediction. We introduce representations, algorithms, machine learning models, and datasets to address the chal- lenges associated with each of these tasks. We introduces new models for 3D semantic segmentation and shape completion. It also introduces a graph representation and search algorithm that facilitates generating multiple variations of a given shape. Last, it contributes a large scale dataset of procedu- ral materials along with a neural network model that generates procedural materials from images. The results of this thesis is a set of methods addressing the aforementioned tasks and supporting the creation of realistic virtual environments at scale.
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 | Tchapmi Petse, Lyne |
---|---|
Degree supervisor | Savarese, Silvio |
Thesis advisor | Savarese, Silvio |
Thesis advisor | Bambos, Nicholas |
Thesis advisor | Pauly, John (John M.) |
Degree committee member | Bambos, Nicholas |
Degree committee member | Pauly, John (John M.) |
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 | Lyne P. Tchapmi. |
---|---|
Note | Submitted to the Department of Electrical Engineering. |
Thesis | Thesis Ph.D. Stanford University 2023. |
Location | https://purl.stanford.edu/vx419pw4129 |
Access conditions
- Copyright
- © 2023 by Lyne Tchapmi Petse
- License
- This work is licensed under a Creative Commons Attribution Non Commercial No Derivatives 3.0 Unported license (CC BY-NC-ND).
Also listed in
Loading usage metrics...