TL;DR本文提出了一种新颖的视频修复方法,使用 Deep Flow Completion 网络综合光流场来引导像素填充,从而保证了视频内容的时空连贯性,并在 DAVIS 和 YouTube-VOS 数据集上取得了最优的修复质量和速度。
Abstract
video inpainting, which aims at filling in missing regions of a video,
remains challenging due to the difficulty of preserving the precise spatial and
temporal coherence of video contents. In this work we propose
Flow-Guided Diffusion model significantly enhances temporal consistency and inpainting quality in video inpainting by employing optical flow and a model-agnostic flow-guided latent interpolation technique.