논문 ★초간단 5줄 요약★ ArcFace: Additive Angular Margin Loss for Deep Face Recognition
ArcFace 논문의 목적
얼굴 인식 성능 높이기
5줄 요약
1. Additive Angular Margin Loss Term 추가
2. 네트워크를 거쳐 도출된 feature embedding(vector)은 정규화를 통해 같은 hyperspher상에 위치.
3. Inter-class 끼리는 feature vector의 geodesic distance 거리가 멀게,
Intra-class 끼리는 sub-class로 더 구분하여 sub-class끼리의 거리가 멀게 학습.
4. Sub-class 중 non-dominant cluster는 noise data로 간주하여 제거.
5. 남은 데이터로 scratch부터 모델 재학습 → 성능 상승
https://arxiv.org/abs/1801.07698
ArcFace: Additive Angular Margin Loss for Deep Face Recognition
Recently, a popular line of research in face recognition is adopting margins in the well-established softmax loss function to maximize class separability. In this paper, we first introduce an Additive Angular Margin Loss (ArcFace), which not only has a cle
arxiv.org
