Unverified50% confidenceBenchmarkExact time
NIST 2019年的测试发现某些面部识别算法对非裔面孔的误识率是白人面孔的10到100倍
1
Sources
50%
Confidence
Long-term
Relevance
8/15/2026
First Seen
Sources
Related Claims
Unverified主流文生图模型的训练数据集（如LAION-5B）中不同族裔面孔分布极不均衡，白人面孔占绝对多数67% similarUnverified面部年龄估算技术在不同种族群体中的准确率存在显著差异65% similarUnverifiedFacial age estimation technology shows significant accuracy disparities across different ethnic groups63% similarUnverified黑白卡图案的对比度和线条粗细比图案丰富度更重要：新生儿视力仅0.01-0.02，需要粗线条（线宽≥1cm）和纯黑纯白（无灰阶过渡）的高对比图案，花哨的多色渐变图案对早期视觉刺激无效反而干扰58% similarUnverified面部表情图案应采用写实卡通脸（明确的眉眼嘴变化）而非抽象符号或纯色块，3-5岁儿童对眉毛角度和嘴角弧度的识别准确率显著高于颜色隐喻（如'红色代表生气'需额外教学成本）57% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/753955API
curl https://kongchang.com/api/v1/knowledge/claims/753955MCP
get_claim(id=753955)