FR Loss on Mnist

Implement diffrent Face Recognition Loss with Pytorch. Testing and visualizing on Mnist. It's just a toy example but very useful to understand and compare those loss function.

Recent Updata

2020.1.14:

• Fix some bugs in ArcFace
• Visualize test data rather than training data
• Add support for our QAMFace Loss

Example

Raw

2D Embedded Feature Normalized Feature

CenterLoss

SphereFace

Quick Start

Dependencies

• Pytorch >=1.0 (0.4 maybe work either)
• tensorboardX >=1.4

I highly recommend you to use Anaconda.

How to run

• Net.py include the implementation of network and loss functions
• train.py contains the training and test process. Editing name to choose a proper loss function.

About the project

• I try to use the same structure to implement different loss. If you have any questions or you find any mistakes, please submmit an issue. Thanks a lot!