서브메뉴
검색
Exploring Knowledge in Generative Models
Exploring Knowledge in Generative Models
상세정보
- 자료유형
- 학위논문 서양
- 최종처리일시
- 20260202105704
- ISBN
- 9798263308124
- DDC
- 004
- 저자명
- Bhattad, Anand.
- 서명/저자
- Exploring Knowledge in Generative Models
- 발행사항
- [Sl] : University of Illinois at Urbana-Champaign, 2024
- 발행사항
- Ann Arbor : ProQuest Dissertations & Theses, 2024
- 형태사항
- 93 p
- 주기사항
- Source: Dissertations Abstracts International, Volume: 87-05, Section: B.
- 주기사항
- Advisor: Forsyth, David A.
- 학위논문주기
- Thesis (Ph.D.)--University of Illinois at Urbana-Champaign, 2024.
- 초록/해제
- 요약Generative models, such as StyleGAN, have demonstrated remarkable ability in producing realistic and controllable images. However, the underlying representations and mechanisms employed by these models remain largely unexplored. This thesis delves into the intrinsic properties and manipulability of StyleGAN, focusing on image relighting and decomposition. We begin by exploring the impact of image decompositions on image-based relighting. By analyzing the role of intrinsic image components such as reflectance, shading, and normals, we gain insights into the fundamental properties that contribute to realistic relighting. This understanding lays the foundation for our subsequent investigations into StyleGAN. Building upon these insights, we introduce StyLitGAN, a method that enables StyleGAN to generate scenes with novel lighting conditions. StyLitGAN produces realistic lighting effects, including cast shadows, soft shadows, inter-reflections, and glossy effects, without requiring labeled, paired, or CGI data. Moreover, it seamlessly extends to manipulating surface properties like colors and materials. Next, we present Make It So, a near-perfect GAN inversion technique that significantly outperforms previous state-of-the-art methods. Make It So can invert and relight real scenes, including out-of-domain images, demonstrating its generalizability and robustness. Finally, we uncover hidden gems within StyleGAN, providing strong evidence that it encodes easily accessible and accurate internal representations of familiar scene properties, known as ``intrinsic images," as defined by Barrow and Tenenbaum in their seminal work from 1978. We demonstrate that StyleGAN has encodings for intrinsic images such as reflectance, shading, and normals, which can be extracted and manipulated for various applications. Through our discoveries, we shed light on the implicit understanding of worldly knowledge present within generative models like StyleGAN. Our findings pave the way for improved manipulability, understanding, and refinement of generative models, with potential applications in computer vision, computational photography, computer graphics, and machine learning. This thesis contributes to the broader goal of leveraging generative models for advanced image manipulation and scene understanding tasks.
- 일반주제명
- Computer science
- 일반주제명
- Computer engineering
- 일반주제명
- Systems science
- 키워드
- Visual knowledge
- 키워드
- Intrinsic images
- 키워드
- Depth
- 키워드
- Albedo
- 키워드
- Segmentation
- 기타저자
- University of Illinois at Urbana-Champaign Computer Science
- 기본자료저록
- Dissertations Abstracts International. 87-05B.
- 전자적 위치 및 접속
- 로그인 후 원문을 볼 수 있습니다.
MARC
008260126s2024 us c eng d■001000017361086
■00520260202105704
■006m o d
■007cr#unu||||||||
■020 ▼a9798263308124
■035 ▼a(MiAaPQ)AAI32409900
■035 ▼a(MiAaPQ)124347
■040 ▼aMiAaPQ▼cMiAaPQ
■0820 ▼a004
■1001 ▼aBhattad, Anand.
■24510▼aExploring Knowledge in Generative Models
■260 ▼a[Sl]▼bUniversity of Illinois at Urbana-Champaign▼c2024
■260 1▼aAnn Arbor▼bProQuest Dissertations & Theses▼c2024
■300 ▼a93 p
■500 ▼aSource: Dissertations Abstracts International, Volume: 87-05, Section: B.
■500 ▼aAdvisor: Forsyth, David A.
■5021 ▼aThesis (Ph.D.)--University of Illinois at Urbana-Champaign, 2024.
■520 ▼aGenerative models, such as StyleGAN, have demonstrated remarkable ability in producing realistic and controllable images. However, the underlying representations and mechanisms employed by these models remain largely unexplored. This thesis delves into the intrinsic properties and manipulability of StyleGAN, focusing on image relighting and decomposition. We begin by exploring the impact of image decompositions on image-based relighting. By analyzing the role of intrinsic image components such as reflectance, shading, and normals, we gain insights into the fundamental properties that contribute to realistic relighting. This understanding lays the foundation for our subsequent investigations into StyleGAN. Building upon these insights, we introduce StyLitGAN, a method that enables StyleGAN to generate scenes with novel lighting conditions. StyLitGAN produces realistic lighting effects, including cast shadows, soft shadows, inter-reflections, and glossy effects, without requiring labeled, paired, or CGI data. Moreover, it seamlessly extends to manipulating surface properties like colors and materials. Next, we present Make It So, a near-perfect GAN inversion technique that significantly outperforms previous state-of-the-art methods. Make It So can invert and relight real scenes, including out-of-domain images, demonstrating its generalizability and robustness. Finally, we uncover hidden gems within StyleGAN, providing strong evidence that it encodes easily accessible and accurate internal representations of familiar scene properties, known as ``intrinsic images," as defined by Barrow and Tenenbaum in their seminal work from 1978. We demonstrate that StyleGAN has encodings for intrinsic images such as reflectance, shading, and normals, which can be extracted and manipulated for various applications. Through our discoveries, we shed light on the implicit understanding of worldly knowledge present within generative models like StyleGAN. Our findings pave the way for improved manipulability, understanding, and refinement of generative models, with potential applications in computer vision, computational photography, computer graphics, and machine learning. This thesis contributes to the broader goal of leveraging generative models for advanced image manipulation and scene understanding tasks.
■590 ▼aSchool code: 0090.
■650 4▼aComputer science
■650 4▼aComputer engineering
■650 4▼aSystems science
■653 ▼aVisual knowledge
■653 ▼aGenerative models
■653 ▼aIntrinsic images
■653 ▼aImage decomposition
■653 ▼aDepth
■653 ▼aAlbedo
■653 ▼aSegmentation
■690 ▼a0984
■690 ▼a0464
■690 ▼a0790
■71020▼aUniversity of Illinois at Urbana-Champaign▼bComputer Science.
■7730 ▼tDissertations Abstracts International▼g87-05B.
■790 ▼a0090
■791 ▼aPh.D.
■792 ▼a2024
■793 ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17361086▼nKERIS▼z이 자료의 원문은 한국교육학술정보원에서 제공합니다.


