Introduction

• [X] InsightFace inference example (production ready architecture)

• [X] Face recognition demo with insightface (visualization missing, add later)

• [ ] InsightFace training pipeline

What does this do if I know nothing about face recognitions?

This is a server, wrapping up with a frozen model, accepting a photo of face, then output a vector of 512 dimension to describe it.

It means:

• [X] You pbbly need another pipeline before this to detect a face bounding box first

• [X] Then you can run this project to describe this face

• [X] Later on, it's up to your purpose, if your purpose is face alignment/detection/distinguish, you need another classifier after this to do the job

An example of this is as following, borrowed from openface:

Demo Facial Recognition App

Yes Yes Yes, I know your are lazy. So I made a demo app for you with following architecture:

How to run it

• Install Depenencies: pip install -r requirements.txt

• Download pre-trained frozen model and put it under pretrained folder

• Run example: python apps/example.py

• You shall be able to see terminal output a 512 element array representing face feature embedded

• Run demo: python apps/demo.py

• You sahll be able to see it output the architecture described above logs

References

• Deng, Jiankang, Jia Guo, and Stefanos Zafeiriou. "Arcface: Additive angular margin loss for deep face recognition." arXiv preprint arXiv:1801.07698 (2018).

• Official Implementation (mxnet): deepinsight/insightface

• Third Party Implementation (tensorflow): auroua/InsightFace_TF