New zero-shot video restoration method uses image diffusion priors

Researchers have developed a new method for zero-shot video restoration by leveraging image diffusion models. This approach tackles the challenge of computational cost by using pre-trained image models instead of training video-specific ones. The key innovation lies in generating initial noise directly from corrupted video frames via DDIM inversion, preserving temporal correlations, and employing adaptive noise blending and trajectory attention to enhance temporal consistency and reduce degradation artifacts. Experiments show this method outperforms existing zero-shot techniques on various video restoration datasets. AI

IMPACT This research could lead to more efficient and effective video restoration tools by leveraging existing image diffusion models.

RANK_REASON The item is an academic paper detailing a new method for video restoration. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• alphaXiv
• arXiv
• CatalyzeX
• DagsHub
• Denoising Diffusion Implicit Models
• Gotit.pub
• Hugging Face
• REDS4
• ScienceCast
• University of California, Davis

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →