메뉴 건너뛰기
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)
고객센터 ENG
주제분류

논문 기본 정보

저자정보
(동아대학교) (동아대학교) (동아대학교) (동아대학교) (동아대학교)
저널정보
한국멀티미디어학회 멀티미디어학회논문지 멀티미디어학회논문지 제29권 제3호
오류 신고하기

피인용 0

검색

    초록·키워드

    Single-image 3D reconstruction remains challenging due to unreliable geometry and inconsistent camera pose estimation. We introduce a diffusion-based pipeline that converts a single input image into a geometrically consistent pseudo-multiview set suitable for stable 3D Gaussian Splatting. Although SV3D synthesizes diverse viewpoints its outputs often contain geometric distortions and unstable view trajectories. To address these issues we apply a three-stage preprocessing and geometric initialization pipeline: (1) appearance-consistency–based filtering along the synthesized view trajectory to select structurally reliable frames, (2) a super-resolution step that restores high-frequency texture details, and (3) VGGT-guided geometric initialization that enables robust COLMAP pose estimation. This pipeline transforms raw SV3D sequences into a coherent pseudo-multiview dataset with stable viewpoints, enhanced textures, and consistent geometric cues. As a result, our method achieves more reliable camera registration and improved reconstruction quality in terms of geometric consistency and visual stability compared to existing single-view approaches relying on strong prior-driven geometry.

    최근 본 자료 전체보기

      UCI(KEPA) : I410-151-26-02-096577225