Unverified50% confidenceFactExact time
人脸识别中的特征向量通常是128维或512维的浮点数向量
1
Sources
50%
Confidence
Long-term
Relevance
8/7/2026
First Seen
Sources
Related Claims
UnverifiedFacial recognition systems used by U.S. federal agencies rely primarily on convolutional neural networks (CNNs), which convert face images into high-dimensional embedding vectors—typically 128 or 512 dimensions—for similarity comparison.79% similarUnverifiedFace feature extraction encodes each face into a high-dimensional vector such as 128-dimensional or 512-dimensional74% similarUnverifiedMediaPipe's face detection module outputs 468 facial landmark coordinates for precise face region localization70% similarUnverified端到端可微的人脸相似度链路包括人脸检测（RetinaFace/MTCNN）、人脸对齐（Spatial Transformer Network）、特征提取（ArcFace/FaceNet/AdaFace，输出512维或128维嵌入）和余弦相似度计算68% similarUnverified外观特征嵌入通过深度神经网络将目标图像块压缩映射为低维稠密向量，通常128维至512维67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/705628API
curl https://kongchang.com/api/v1/knowledge/claims/705628MCP
get_claim(id=705628)