Additive Margin Softmax for Face Verification

Abstract

In this paper, we propose a conceptually simple and geometrically interpretable objective function, i.e. additive margin Softmax (AM-Softmax), for deep face verification. In general, the face verification task can be viewed as a metric learning problem, so learning large-margin face features whose intra-class variation is small and inter-class difference is large is of great importance in order to achieve good performance. Recently, Large-margin Softmax and Angular Softmax have been proposed to incorporate the angular margin in a multiplicative manner. In this work, we introduce a novel additive angular margin for the Softmax loss, which is intuitively appealing and more interpretable than the existing works. We also emphasize and discuss the importance of feature normalization in the paper. Most importantly, our experiments on LFW BLUFR and MegaFace show that our additive margin softmax loss consistently performs better than the current state-of-the-art methods using the same network architecture and training dataset. Our code has also been made available at https://github.com/happynear/AMSoftmax

Published in Signal Processing Letters, Volume: 25 Issue: 7 Pages: 926-930

References in corpus (3)

• FaceNet: A Unified Embedding for Face Recognition and Clustering
• Joint Face Detection and Alignment using Multi-task Cascaded Convolutional Networks
• NormFace: L2 Hypersphere Embedding for Face Verification

Cited by in corpus (50)

• Deep Face Recognition: A Survey
• AdvHat: Real-world adversarial attack on ArcFace Face ID system
• One-class Learning Towards Synthetic Voice Spoofing Detection
• SphereReID: Deep Hypersphere Manifold Embedding for Person Re-Identification
• Analyzing Overfitting under Class Imbalance in Neural Networks for Image Segmentation
• On Low-Resolution Face Recognition in the Wild: Comparisons and New Techniques
• Representation Learning by Rotating Your Faces
• Recent Advances in Deep Learning Techniques for Face Recognition
• Semantic Models for the First-stage Retrieval: A Comprehensive Review
• Efficient Facial Representations for Age, Gender and Identity Recognition in Organizing Photo Albums using Multi-output CNN
• Cross-Resolution Learning for Face Recognition
• Learning to Disentangle Scenes for Person Re-identification
• Deep Feature Space: A Geometrical Perspective
• Towards NIR-VIS Masked Face Recognition
• Minimum Margin Loss for Deep Face Recognition
• AM-MobileNet1D: A Portable Model for Speaker Recognition
• Entropic Out-of-Distribution Detection: Seamless Detection of Unknown Examples
• Survey on the Analysis and Modeling of Visual Kinship: A Decade in the Making
• MGH: Metadata Guided Hypergraph Modeling for Unsupervised Person Re-identification
• Entropic Out-of-Distribution Detection
• Regular Polytope Networks
• Attention Back-end for Automatic Speaker Verification with Multiple Enrollment Utterances
• Learning Imbalanced Datasets with Maximum Margin Loss
• The VoxCeleb Speaker Recognition Challenge: A Retrospective
• A Unified Deep Learning Framework for Short-Duration Speaker Verification in Adverse Environments
• Data-Free/Data-Sparse Softmax Parameter Estimation with Structured Class Geometries
• Why do Angular Margin Losses work well for Semi-Supervised Anomalous Sound Detection?
• Additive Margin SincNet for Speaker Recognition
• Channel adversarial training for speaker verification and diarization
• Adjusting Logit in Gaussian Form for Long-Tailed Visual Recognition
• Rapid detection and recognition of whole brain activity in a freely behaving Caenorhabditis elegans
• Person Recognition in Personal Photo Collections
• Building Computationally Efficient and Well-Generalizing Person Re-Identification Models with Metric Learning
• Generalizing Speaker Verification for Spoof Awareness in the Embedding Space
• Toward Improving Synthetic Audio Spoofing Detection Robustness via Meta-Learning and Disentangled Training With Adversarial Examples
• Leveraging Angular Distributions for Improved Knowledge Distillation
• A Softmax-free Loss Function Based on Predefined Optimal-distribution of Latent Features for Deep Learning Classifier
• Cluster-Guided Unsupervised Domain Adaptation for Deep Speaker Embedding
• Cosine Scoring with Uncertainty for Neural Speaker Embedding
• Tackling the Score Shift in Cross-Lingual Speaker Verification by Exploiting Language Information
• Open-set Face Recognition using Ensembles trained on Clustered Data
• Prototype Memory for Large-scale Face Representation Learning
• Curricular SincNet: Towards Robust Deep Speaker Recognition by Emphasizing Hard Samples in Latent Space
• A t-distribution based operator for enhancing out of distribution robustness of neural network classifiers
• Adaptive Sparse Softmax: An Effective and Efficient Softmax Variant
• ESANS: Effective and Semantic-Aware Negative Sampling for Large-Scale Retrieval Systems
• A Hybrid Cross-Stage Coordination Pre-ranking Model for Online Recommendation Systems
• CAMRI Loss: Improving Recall of a Specific Class without Sacrificing Accuracy
• Hyperbolic Additive Margin Softmax with Hierarchical Information for Speaker Verification
• Cross-Corpora Spoken Language Identification with Domain Diversification and Generalization