Efficient optimization via approximation
Abstract/Contents
- Abstract
- Optimization in nonrigid registration and image processing is often non-convex and computationally expensive. To address these challenges, I will present approaches for efficient optimization via approximation. First, I will show that classical models can achieve state-of-the-art performance with approximate global optimization, for the problems of nonrigid 3D registration and optical flow. We cast the objective as discrete Markov random field optimization and apply efficient global optimization algorithms based on linear programming relaxation. For nonrigid 3D registration, our approach outperforms a large body of prior work by a significant margin, increasing registration precision on real data by a factor of 3. For optical flow, we show that one-shot global optimization of a classical Horn-Schunck-type objective over regular grids at a single resolution is sufficient to initialize continuous interpolation and achieve state-of-the-art performance on challenging modern benchmarks. Second, I will present an approximation approach to accelerating a wide variety of image processing operators. We show that convolutional networks can model the action of many image processing operators more accurately than even sophisticated recent downsample-evaluate-upsample schemes while being up to an order of magnitude faster.
Description
Type of resource | text |
---|---|
Form | electronic; electronic resource; remote |
Extent | 1 online resource. |
Publication date | 2017 |
Issuance | monographic |
Language | English |
Creators/Contributors
Associated with | Chen, Qifeng |
---|---|
Associated with | Stanford University, Computer Science Department. |
Primary advisor | James, Doug |
Primary advisor | Koltun, Vladlen, 1980- |
Thesis advisor | James, Doug |
Thesis advisor | Koltun, Vladlen, 1980- |
Thesis advisor | Ermon, Stefano |
Advisor | Ermon, Stefano |
Subjects
Genre | Theses |
---|
Bibliographic information
Statement of responsibility | Qifeng Chen. |
---|---|
Note | Submitted to the Department of Computer Science. |
Thesis | Thesis (Ph.D.)--Stanford University, 2017. |
Location | electronic resource |
Access conditions
- Copyright
- © 2017 by Qifeng Chen
- 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...