서브메뉴
검색
Learning Domain-Specific Cameras.
Learning Domain-Specific Cameras.
상세정보
- 자료유형
- 학위논문
- Control Number
- 0017162990
- International Standard Book Number
- 9798384426554
- Dewey Decimal Classification Number
- 004
- Main Entry-Personal Name
- Nguyen, Cindy My Anh.
- Publication, Distribution, etc. (Imprint
- [S.l.] : Stanford University., 2024
- Publication, Distribution, etc. (Imprint
- Ann Arbor : ProQuest Dissertations & Theses, 2024
- Physical Description
- 119 p.
- General Note
- Source: Dissertations Abstracts International, Volume: 86-04, Section: B.
- General Note
- Advisor: Wetzstein, Gordon.
- Dissertation Note
- Thesis (Ph.D.)--Stanford University, 2024.
- Summary, Etc.
- 요약Many autonomous systems rely on images taken using off-the-shelf cameras. These images are passed to a domain-specific algorithm that provides high-level predictions, such as classification or segmentation. However, general digital cameras are optimized for aesthetics and may not necessarily provide the most useful information for downstream tasks. Additionally, these autonomous pipelines often have narrow use cases, performing only one or a few tasks with the captured images. We explore how to extract the most performance in such cases by learning end-to-end domain-specific camera pipelines.To learn a domain-specific camera, we can learn the parameters of each stage of the pipeline, such as the optics, sensors, and software. We built and evaluated several domain-specific camera systems that demonstrate the benefits of this paradigm. First, we demonstrate how optimizing both lens design and software can lead to better performance on monocular depth estimation. Next, we describe a system in which we jointly optimize the sensor and a reconstruction network to perform single-snapshot motion deblurring. Finally, we focus on optimizing the algorithmic part of a pipeline to bring robust performance to low-light text recognition using diffusion models.
- Subject Added Entry-Topical Term
- Computer science.
- Subject Added Entry-Topical Term
- Optics.
- Added Entry-Corporate Name
- Stanford University.
- Host Item Entry
- Dissertations Abstracts International. 86-04B.
- Electronic Location and Access
- 로그인을 한후 보실 수 있는 자료입니다.
- Control Number
- joongbu:657691
MARC
008250224s2024 us ||||||||||||||c||eng d■001000017162990
■00520250211152121
■006m o d
■007cr#unu||||||||
■020 ▼a9798384426554
■035 ▼a(MiAaPQ)AAI31460389
■035 ▼a(MiAaPQ)Stanfordzm136ny2176
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a004
■1001 ▼aNguyen, Cindy My Anh.
■24510▼aLearning Domain-Specific Cameras.
■260 ▼a[S.l.]▼bStanford University. ▼c2024
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2024
■300 ▼a119 p.
■500 ▼aSource: Dissertations Abstracts International, Volume: 86-04, Section: B.
■500 ▼aAdvisor: Wetzstein, Gordon.
■5021 ▼aThesis (Ph.D.)--Stanford University, 2024.
■520 ▼aMany autonomous systems rely on images taken using off-the-shelf cameras. These images are passed to a domain-specific algorithm that provides high-level predictions, such as classification or segmentation. However, general digital cameras are optimized for aesthetics and may not necessarily provide the most useful information for downstream tasks. Additionally, these autonomous pipelines often have narrow use cases, performing only one or a few tasks with the captured images. We explore how to extract the most performance in such cases by learning end-to-end domain-specific camera pipelines.To learn a domain-specific camera, we can learn the parameters of each stage of the pipeline, such as the optics, sensors, and software. We built and evaluated several domain-specific camera systems that demonstrate the benefits of this paradigm. First, we demonstrate how optimizing both lens design and software can lead to better performance on monocular depth estimation. Next, we describe a system in which we jointly optimize the sensor and a reconstruction network to perform single-snapshot motion deblurring. Finally, we focus on optimizing the algorithmic part of a pipeline to bring robust performance to low-light text recognition using diffusion models.
■590 ▼aSchool code: 0212.
■650 4▼aComputer science.
■650 4▼aOptics.
■690 ▼a0752
■690 ▼a0984
■71020▼aStanford University.
■7730 ▼tDissertations Abstracts International▼g86-04B.
■790 ▼a0212
■791 ▼aPh.D.
■792 ▼a2024
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17162990▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.
미리보기
내보내기
chatGPT토론
Ai 추천 관련 도서
Buch Status
- Reservierung
- 캠퍼스간 도서대출
- 서가에 없는 책 신고
- Meine Mappe
