# SphereFace-1

**Repository Path**: nuaacj/SphereFace-1

## Basic Information

- **Project Name**: SphereFace-1
- **Description**: 🍑 TensorFlow Code for CVPR 2017 paper "SphereFace: Deep Hypersphere Embedding for Face Recognition"
- **Primary Language**: Unknown
- **License**: MIT
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 1
- **Forks**: 0
- **Created**: 2020-11-10
- **Last Updated**: 2021-11-03

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

[![Build Status](https://travis-ci.org/raghakot/keras-resnet.svg?branch=master)](https://github.com/YunYang1994/SphereFace)
[![license](https://img.shields.io/github/license/mashape/apistatus.svg?maxAge=2592000)](https://github.com/YunYang1994/SphereFace/blob/master/LICENSE)

|Author|YunYang1994|
|---|---
|E-mail|dreameryangyun@sjtu.edu.cn

SphereFace
===========================
This is a quick implementation for [Deep Hypersphere Embedding for Face Recognition](https://arxiv.org/abs/1704.08063)(CVPR 2017).This paper proposed the angular softmax loss that enables convolutional neural networks(CNNs) to learn angularly discriminative features. The main content I replicated contains: <br>

- **1**. mathematical comparison among original softmax, modified softmax and angular softmax;
- **2**. show the accuracy and loss comparison of different softmax in the experiment;
- **3**. 2D and 3D visualization of embeddings learned with different softmax loss on MNIST dataset;
- **4**. I replicated [SphereFace-20](http://ethereon.github.io/netscope/#/gist/20f6ddf70a35dec5019a539a502bccc5) model with tensorflow.and here is [code](https://github.com/YunYang1994/SphereFace/blob/master/model.py) and [graph](https://github.com/YunYang1994/SphereFace/blob/master/image/model_graph.png).

### CNN-network
many current CNNS can viewed as convolution feature learning guided by softmax loss on top. however, softmax is easy to to optimize but does not explicitly encourage large margin between different classes.
<p align="center">
    <img width="70%" src="https://github.com/YunYang1994/SphereFace/blob/master/image/network.png" style="max-width:90%;">
    </a>
</p>
on this situation, the author proposed a new loss function that always encourages an angular decision margin between different classes. 



### softmax loss
|softmax|formula|test acc(MNIST)|
|---|---|:---:|
|original softmax|![weibo-logo](https://github.com/YunYang1994/SphereFace/blob/master/image/original_softmax.png)|0.9775|
|modified softmax|![weibo-logo](https://github.com/YunYang1994/SphereFace/blob/master/image/modified_softmax.png)|0.9847|
|angular softmax|![weibo-logo](https://github.com/YunYang1994/SphereFace/blob/master/image/angular_softmax.png)|0.9896|

A toy example on MNIST dataset, CNN features can be visualized by setting the output dimension as 2 or 3, as shown in following figures.

### 2D visualization
|original softmax|modified softmax|angular softmax|
|---|---|:---:|
|![weibo-logo](https://github.com/YunYang1994/SphereFace/blob/master/image/2D_Original_Softmax_Loss_embeddings.gif)|![weibo-logo](https://github.com/YunYang1994/SphereFace/blob/master/image/2D_Modified_Softmax_Loss_embeddings.gif)|![weibo-logo](https://github.com/YunYang1994/SphereFace/blob/master/image/2D_Angular_Softmax_Loss_embedding.gif)|

### 3D visualization
|original softmax|modified softmax|angular softmax|
|---|---|:---:|
|![weibo-logo](https://github.com/YunYang1994/SphereFace/blob/master/image/3D_Original_Softmax_Loss_embeddings.gif)|![weibo-logo](https://github.com/YunYang1994/SphereFace/blob/master/image/3D_Modified_Softmax_Loss_embeddings.gif)|![weibo-logo](https://github.com/YunYang1994/SphereFace/blob/master/image/3D_Angular_Softmax_Loss_embeddings.gif)|

### loss and accuracy
|training loss|training accuracy|
|---|:---:|
|![weibo-logo](https://github.com/YunYang1994/SphereFace/blob/master/image/train_loss.jpg)|![weibo-logo](https://github.com/YunYang1994/SphereFace/blob/master/image/train_acc.jpg)|