Unverified50% confidenceBenchmarkExact time
CELEBV-HQ是一个包含35,666个高质量人脸视频片段的大规模数据集，是评估视频生成模型长序列保真度的标准基准之一
1
Sources
50%
Confidence
Long-term
Relevance
8/8/2026
First Seen
Sources
往返一致性：让扩散模型自己预测误差的自监督方法
Related Claims
UnverifiedAnti-UAV410 是2023年发布的公开基准数据集，包含410个视频序列、超过438,000帧标注图像，涵盖可见光和红外热成像两个模态61% similarUnverifiedVideo VQ-VAE, 3D convolution, and encoding uniformly sampled keyframes with ViT are the mainstream video tokenization methods.61% similarUnverifiedThe video generation feature in Coze's official templates consumes 5,207 resource points per use.60% similarUnverified视频生成任务的显存占用约为同分辨率图像任务的10–30倍59% similarVerifiedGenerating an 8-second, 24-fps video means the model must maintain semantic consistency across 192 frames simultaneously on the time axis, with a computational load approximately tens of times greater than that of a single image at the same resolution.58% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/707116API
curl https://kongchang.com/api/v1/knowledge/claims/707116MCP
get_claim(id=707116)