wy1iu/sphereface

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

1606 stars · 530 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

Official implementation of SphereFace (Deep Hypersphere Embedding for Face Recognition, CVPR 2017), built on Caffe with a full face recognition pipeline including face detection, alignment, and recognition. It introduces the A-Softmax (angular softmax) loss for learning discriminative angular face embeddings.

Use cases

• train a face recognition model with angular margin loss
• reproduce SphereFace results on MegaFace and LFW benchmarks
• build a face verification pipeline with detection and alignment
• research angular margin loss functions for face embeddings
• extract face embeddings for identity matching
• compare A-Softmax against softmax for face recognition

When to choose

• you need the original reference implementation of SphereFace/A-Softmax for research or paper reproduction
• you work with Caffe and want the complete detection-alignment-recognition pipeline
• you are studying angular margin losses in metric learning

When to avoid

• you want a modern, easy-to-use PyTorch implementation (use OpenSphere instead)
• you need production face recognition with maintained dependencies
• you cannot build Caffe or lack GPU infrastructure

Facets

library · maturity maintenance

deep-learning computer-vision image-processing machine-learning computer-vision deep-learning image-processing artificial-intelligence python cpp face-recognition sphereface angular-softmax caffe cvpr-2017 face-detection face-alignment research-code gpu linux

Member repositories

For agents

markdown · JSON · MCP: product_card(name="wy1iu/sphereface")