메뉴 건너뛰기
.. 내서재 .. 알림
소속 기관/학교 인증
인증하면 논문, 학술자료 등을  무료로 열람할 수 있어요.
한국대학교, 누리자동차, 시립도서관 등 나의 기관을 확인해보세요
(국내 대학 90% 이상 구독 중)
로그인 회원가입 고객센터 ENG
주제분류

추천
검색

논문 기본 정보

자료유형
학위논문
저자정보

김건우 (포항공과대학교, 포항공과대학교 일반대학원)

지도교수
조성현
발행연도
2023
저작권
포항공과대학교 논문은 저작권에 의해 보호받습니다.

이용수2

표지
AI에게 요청하기
추천
검색

이 논문의 연구 히스토리 (2)

초록· 키워드

오류제보하기
RAW images store physically-meaningful recordings of light intensity that can benefit various computational-photography tasks such as denoising and photo retouching. However, RAW images are rarely shared mainly due to its excessive data size compared to their sRGB counterparts processed by a series of internal image processing operations in camera ISPs. Learning the forward and inverse processes of camera ISPs has been recently demonstrated, enabling RAW-level image processing on sRGB images. However, existing learning-based ISP methods fail to handle the large variation in the ISP processes of different cameras and their dynamic behaviors with respect to ISP parameters, as they consider neither multiple cameras nor ISP parameters. In this paper, we propose HyperISP, a learning-based method for forward and inverse conversion between sRGB and RAW images, that adopts a hypernetwork to model the variation caused by different cameras and ISP parameters. Given a specific camera model and ISP parameters, the hypernetwork modulates the forward and inverse ISP networks to adapt them to the camera and parameters. Extensive experiments demonstrate that HyperISP can successfully adapt to different cameras and ISP parameters, and achieve superior reconstruction results to previous methods.

목차

1. Introduction 1
2. Related Work 4
3. Method 5
3.1 Forward ISP Network 5
3.1.1 Canonical Operation Block 5
3.1.2 LocalNet and GlobalNet 6
3.2 Inverse ISP Network 9
3.3 AdaptNet 9
3.4 Training 11
4. Experiments 13
4.1 Ablation Study 13
4.1.1 HyperISP vs Individual ISPs 13
4.1.2 Ablation Study of the Input Features 15
4.1.3 Ablation Study of the ISP Parameters 16
4.1.4 Ablation Study of the Network Components 18
4.2 Comparison on RAW & sRGB Reconstruction 19
4.2.1 sRGB-to-RAW Reconstruction 20
4.2.2 RAW-to-sRGB Reconstruction 22
4.2.3 Cyclic sRGB Reconstruction 24
4.3 HyperISP for Single Camera Dataset 26
4.4 Camera-to-Camera Transfer 27
4.5 Additional Examples 27
5. Conclusion 34
5.1 Limitation 34
5.2 Future Works 34
Summary (in Korean) 35
References 36

최근 본 자료

전체보기

댓글(0)

0