Unverified50% confidenceOpinionExact time

在深度伪造检测的生产环境中真实图像与伪造图像比例可能高达1000:1甚至更高，极端不平衡下AUC对长尾失败几乎没有区分度

1

Sources

50%

Confidence

Long-term

Relevance

8/27/2026

First Seen

Sources

Deepfake Detectors Fail in the Diffusion Model Era: Why AUC Is No Longer a Reliable Metric

Related Entities

Deepfake Detectionconcept AUCconcept

Related Claims

Unverified在实验室干净输出上表现完美的深度伪造检测器，进入真实内容分发环境经过平台图像处理后可能判别能力急剧下降73% similar Unverified高度写实的AI图像降低了虚假新闻配图、伪造证据和深度伪造内容的门槛69% similar Unverified顶级深度伪造检测模型碰到经过压缩的视频时识别率会大幅下降69% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/809538

API

curl https://kongchang.com/api/v1/knowledge/claims/809538

MCP

get_claim(id=809538)