NormFace: L2 Hypersphere Embedding for Face Verification

Abstract

Thanks to the recent developments of Convolutional Neural Networks, the performance of face verification methods has increased rapidly. In a typical face verification method, feature normalization is a critical step for boosting performance. This motivates us to introduce and study the effect of normalization during training. But we find this is non-trivial, despite normalization being differentiable. We identify and study four issues related to normalization through mathematical analysis, which yields understanding and helps with parameter settings. Based on this analysis we propose two strategies for training using normalized features. The first is a modification of softmax loss, which optimizes cosine similarity instead of inner-product. The second is a reformulation of metric learning by introducing an agent vector for each class. We show that both strategies, and small variants, consistently improve performance by between 0.2% to 0.4% on the LFW dataset based on two models. This is significant because the performance of the two models on LFW dataset is close to saturation at over 98%. Codes and models are released on https://github.com/happynear/NormFace

camera-ready version

Cited by in corpus (36)

• A Survey of Deep Learning-based Object Detection
• Learning Discriminative Features with Multiple Granularities for Person Re-Identification
• Additive Margin Softmax for Face Verification
• Deep Face Recognition: A Survey
• Towards Transferable Adversarial Attack against Deep Face Recognition
• SFace: Sigmoid-Constrained Hypersphere Loss for Robust Face Recognition
• Recent Advances in Deep Learning Techniques for Face Recognition
• Semantic Models for the First-stage Retrieval: A Comprehensive Review
• Circle Loss: A Unified Perspective of Pair Similarity Optimization
• SCAN: Self-and-Collaborative Attention Network for Video Person Re-identification
• Long-tailed Visual Recognition via Gaussian Clouded Logit Adjustment
• Variational Graph Normalized Auto-Encoders
• Few-Shot Learning with Geometric Constraints
• Meta Balanced Network for Fair Face Recognition
• Learning to Disentangle Scenes for Person Re-identification
• CALM: Conditional Adversarial Latent Models for Directable Virtual Characters
• Relational Deep Feature Learning for Heterogeneous Face Recognition
• Feature Fusion from Head to Tail for Long-Tailed Visual Recognition
• Distilling Knowledge by Mimicking Features
• Deep Representation Learning on Long-tailed Data: A Learnable Embedding Augmentation Perspective
• Class-incremental Learning with Pre-allocated Fixed Classifiers
• Unbiased organism-agnostic and highly sensitive signal peptide predictor with deep protein language model
• Adap-$\tau$: Adaptively Modulating Embedding Magnitude for Recommendation
• Class Distance Weighted Cross Entropy Loss for Classification of Disease Severity
• Ultra-High-Resolution Detector Simulation with Intra-Event Aware GAN and Self-Supervised Relational Reasoning
• Regular Polytope Networks
• Families In Wild Multimedia: A Multimodal Database for Recognizing Kinship
• Adjusting Logit in Gaussian Form for Long-Tailed Visual Recognition
• Rapid detection and recognition of whole brain activity in a freely behaving Caenorhabditis elegans
• CATFace: Cross-Attribute-Guided Transformer with Self-Attention Distillation for Low-Quality Face Recognition
• Prototype Memory for Large-scale Face Representation Learning
• Bringing Online Egocentric Action Recognition into the wild
• Annotation-Free Human Sketch Quality Assessment
• Joint Discriminative and Metric Embedding Learning for Person Re-Identification
• Open-Set Face Identification on Few-Shot Gallery by Fine-Tuning
• Spherical Feature Transform for Deep Metric Learning