Unverified50% confidenceFactExact time
FVD是FID在视频领域的扩展版本，在计算分布差异时考虑了时间维度的特征
1
Sources
50%
Confidence
Long-term
Relevance
8/7/2026
First Seen
Sources
AI视频生成进化史：从Will Smith吃面翻车到电影级画质
Related Claims
VerifiedFVD was proposed by Unterthiner et al. in 2019; it uses an I3D network pretrained on the Kinetics dataset to extract spatiotemporal video features and computes the Fréchet distance between the distributions of real and generated videos.72% similarUnverifiedCurrent mainstream text-to-video models are generally based on diffusion model architectures, generating coherent video frame sequences by extending image generation capability along the temporal dimension.64% similarUnverified2023年后，以Stable Video Diffusion、Sora为代表的时序扩散模型开始将视频理解为时空连续体，能够在潜空间中同时建模多帧之间的运动关系63% similarUnverified均匀采样按固定时间间隔抽帧，实现简单且计算开销可预测，但对视频内容一无所知，容易错失信息密度不均视频的关键瞬间62% similarUnverified传统视频处理通过光流估计（Optical Flow）和运动向量（Motion Vector）来维护时间连续性62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/700782API
curl https://kongchang.com/api/v1/knowledge/claims/700782MCP
get_claim(id=700782)