---
tags: 生物辨識
---

# Variational Prototype Learning for Deep Face Recognition

## Introduction

#### sample-to-sample

facenet，triplet loss需要建立triplet-pair

#### sample-to-prototype

cosface,arcface....，從softmax loss變化的一系列方法


## Method

#### VPL framework

we train the networks for several epochs, until the features are stable across hundred of steps


#### loss function



#### Backpropagation



## Experiments