Learning and exploring the compositional structure of 3D data
Abstract/Contents
- Abstract
- 3D data arising either from scanning real objects with depth sensors or from modeling by designers have a unique characteristic — they represent the entire geometry of objects in a real world. As such, unlike 2D images containing only a projected view with occlusions and clutter, they directly enable understanding of how the objects are composed and structured in an actual physical space. In this dissertation, we present novel methodologies for learning the compositional structure of 3D data from a collection and study various downstream applications, enabled by this structure extraction. The main contributions are organized into three parts: shape decomposition, structure knowledge aggregation, and shape composition. In the decomposition part, we first discuss ideas for parsing raw scanned 3D data and segmenting them into multiple parts based on either their local geometry or topological structure. For solving the model estimation problems more robustly, we propose frameworks integrating supervised learning and optimization techniques, taking advantage of supervision while guaranteeing local optimality. In the knowledge aggregation part, we consider the case when the compositional structure (e.g., parts or keypoints) given for each 3D shape is inconsistent across the collection due to the difference in data sources. We introduce a neural network that can canonicalize the information defined on independent shapes, without any supervision through correspondences. Lastly, in the composition part, we show how a part-based representation of 3D objects can facilitate shape creation and editing. This representation allows efficient exploration over the shape variation space by fusing a discrete, combinatorial global structure space with a continuous local geometry space. Our methods also learn the part-level structure with unlabeled parts and discover their semantic relations both in a single object and across shapes based on contextual information.
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 | 2019; ©2019 |
| Publication date | 2019; 2019 |
| Issuance | monographic |
| Language | English |
Creators/Contributors
| Author | Sung, Minhyuk |
|---|---|
| Degree supervisor | Guibas, Leonidas J |
| Thesis advisor | Guibas, Leonidas J |
| Thesis advisor | James, Doug L |
| Thesis advisor | Savarese, Silvio |
| Degree committee member | James, Doug L |
| Degree committee member | Savarese, Silvio |
| Associated with | Stanford University, Computer Science Department. |
Subjects
| Genre | Theses |
|---|---|
| Genre | Text |
Bibliographic information
| Statement of responsibility | Minhyuk Sung. |
|---|---|
| Note | Submitted to the Computer Science Department. |
| Thesis | Thesis Ph.D. Stanford University 2019. |
| Location | electronic resource |
Access conditions
- Copyright
- © 2019 by Minhyuk Sung
- License
- This work is licensed under a Creative Commons Attribution Non Commercial 3.0 Unported license (CC BY-NC).
Also listed in
Loading usage metrics...