Unverified50% confidenceFactExact time
对抗样本攻击只需在合成视频上叠加人眼不可见的微小扰动，就能让检测模型准确率从80%跌至接近随机水平
1
Sources
50%
Confidence
Long-term
Relevance
7/18/2026
First Seen
Sources
Related Claims
Unverified对抗样本攻击由Szegedy等人于2013年首次系统性揭示，只需叠加人眼不可见的微小扰动（幅度仅像素值1%~2%）即可让检测模型准确率从80%跌至接近随机水平83% similarUnverified2023年多项学术研究显示在模拟对抗性环境中主流目标检测算法的准确率可下降40-80%69% similarUnverified据部分版权方估计，对于经过精心处理的解说视频，Content ID自动检测的成功率可能不足30%67% similarUnverifiedThe system has a low error rate, with a strange description (such as missing a checkmate) occurring approximately once every 20 videos.66% similarUnverifiedMultimodal large models have extremely low factual recall accuracy in video question answering, which is difficult to improve even with supervised fine-tuning.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/551992API
curl https://kongchang.com/api/v1/knowledge/claims/551992MCP
get_claim(id=551992)