Unverified50% confidenceFactExact time
ArcFace first applies L2 normalization to both the feature embedding vectors and the classification weight matrix, then adds a fixed angular margin (typically 0.5 radians) to the cosine of the angle, scales the result, and feeds it into Softmax to compute the loss.
1
Sources
50%
Confidence
Long-term
Relevance
10/7/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe core idea of loss functions such as ArcFace and CosFace is to artificially enlarge inter-class margins and compress intra-class distributions in angular space to improve discriminability.76% similarUnverifiedArcFace is a deep metric learning method optimized specifically for face recognition that outperforms general visual encoders in face identity feature extraction accuracy by introducing an additive angular margin loss function.74% similarUnverifiedArcFace proposed the Additive Angular Margin Loss in 2018, achieving over 99.8% recognition accuracy on standard benchmarks like LFW73% similarUnverifiedArcFace was proposed by Deng Jiankang et al. in 2019 and employs an additive angular margin loss.69% similarUnverifiedoutline支持outline-offset属性，可以设置负值让轮廓向内收缩63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/983173API
curl https://kongchang.com/api/v1/knowledge/claims/983173MCP
get_claim(id=983173)