FlowVid（CVPR 2024）：通过处理不完美光流实现一致的视频到视频合成

会议：CVPR 2024

发表日期：2023/12/29

资料加入日期：2026-04-14

一句话结论

这篇工作把 optical-flow-aware consistency 路线补进库里，强化了视频到视频合成的一致性技术分支。

问题定义

它要解决的是 video-to-video synthesis 中光流不完美导致的一致性破坏问题。对当前知识库来说，它补的是 RAVE / StableVideo 之外另一条 consistency 技术路线。

方法概述

FlowVid 通过处理 imperfect optical flows 来提升 consistent video-to-video synthesis 的稳定性，使视频变换过程更平滑连贯。

关键发现

它说明视频一致性优化并不只靠扩散噪声和 attention，也可以借助显式运动/流场信息。
它让 consistency 路线从“编辑机制内优化”扩展到“显式时空对齐信号”层。
它有助于以后比较 flow-guided 与 diffusion-native consistency 方法。

局限或疑问

光流本身的不稳定会限制上限。
它更偏一致性工程路线，而不是更高层编辑理解。
它适合作为 consistency 技术分支补层。

原始链接

https://arxiv.org/abs/2312.17681
https://arxiv.org/pdf/2312.17681

备注

FlowVid 在这套库里的作用，是把 flow-guided consistency 路线补成明确节点。

Metadata

{ "id": "2026-04-14-flowvid", "type": "source", "title": "FlowVid（CVPR 2024）：通过处理不完美光流实现一致的视频到视频合成", "status": "reviewed", "created": "2026-04-14", "updated": "2026-04-15", "venue": "CVPR 2024", "ingested_at": "2026-04-14", "tags": [ "near-cvpr-2025", "video-editing", "video-generation", "primary-source" ], "note_status": "reviewed", "source_type": "paper", "authors": [ "Liang, Feng", "Wu, Bichen", "Wang, Jialiang", "Yu, Licheng", "Li, Kunpeng", "Zhao, Yinan", "Misra, Ishan", "Huang, Jia-Bin", "Zhang, Peizhao", "Vajda, Peter", "Marculescu, Diana" ], "published_at": "2023/12/29", "canonical_links": [ "https://arxiv.org/abs/2312.17681", "https://arxiv.org/pdf/2312.17681" ], "raw_entry": "raw/ingest/2026-04-14-flowvid/", "topics": [ "topics/video-editing", "topics/video-generation" ], "entities": [ "entities/video-editing-understanding" ], "claims": [], "questions": [ "questions/question-do-benchmarks-track-real-video-editing-understanding" ] }