인문학
사회과학
자연과학
공학
의약학
농수해양학
예술체육학
복합학
DBpia AI
AI 에이전트
AI 아이디어
AI 리더
AI 뷰어
Citeasy
크롬 서지관리 다운로드
지원사업
학술연구/단체지원/교육 등 연구자 활동을 지속하도록 DBpia가 지원하고 있어요.
커뮤니티
연구자들이 자신의 연구와 전문성을 널리 알리고, 새로운 협력의 기회를 만들 수 있는 네트워킹 공간이에요.
피인용 0
초록·키워드
Recent advances in diffusion-based generative models have enabled smooth and realistic image synthesis, yet image morphing, creating temporally and semantically consistent transitions between two images, remains challenging without task-specific training. Existing approaches often rely on either geometric warping or feature blending, which struggle to maintain semantic coherence and structural fidelity, particularly when the input images differ significantly in appearance or content.
To address these limitations, we propose DualMorph, a training-free image morphing framework that leverages dual DDIM inversion. The method performs independent DDIM inversion on two input images to extract their latent representations, interpolates them within the diffusion latent space, and employs the interpolated latents and text embeddings in the DDIM denoising process to generate intermediate frames. This dual-inversion and latent-space interpolation strategy allows the model to preserve each input’s structural integrity while smoothly merging their semantic attributes, resulting in more coherent and temporally stable morphing transitions without any additional training.
Experiments on public morphing benchmarks demonstrate that DualMorph achieves smoother transitions and improved perceptual similarity compared to existing training-free methods. These results highlight the effectiveness of utilizing latent- space interpolation for continuous and visually coherent image morphing without additional training.
To address these limitations, we propose DualMorph, a training-free image morphing framework that leverages dual DDIM inversion. The method performs independent DDIM inversion on two input images to extract their latent representations, interpolates them within the diffusion latent space, and employs the interpolated latents and text embeddings in the DDIM denoising process to generate intermediate frames. This dual-inversion and latent-space interpolation strategy allows the model to preserve each input’s structural integrity while smoothly merging their semantic attributes, resulting in more coherent and temporally stable morphing transitions without any additional training.
Experiments on public morphing benchmarks demonstrate that DualMorph achieves smoother transitions and improved perceptual similarity compared to existing training-free methods. These results highlight the effectiveness of utilizing latent- space interpolation for continuous and visually coherent image morphing without additional training.
본문·목차
인공지능 문자 인식 모델을 통해 추출된 텍스트로, 일부 오타나 오류가 포함될 수 있으나 지속적으로 개선 중입니다.
오류를 발견하셨다면 해당 부분을 드래그한 후 ' 를 통해 신고해주세요.
오류를 발견하셨다면 해당 부분을 드래그한 후 ' 를 통해 신고해주세요.
최근 본 자료 전체보기
UCI(KEPA) : I410-151-26-02-095552724
