Open source big face: InsightFace takes you to create a face recognition system for unpacking


Referring to the open source technology of artificial intelligence and face recognition, I believe that many peers in the industry should be unfamiliar with InsightFace, which is quoted more than 2100+, since 2018 Open Source, also maintains a strong growth situation in GitHub .

It can also be seen on the actual effect map displayed in the project.

And in the past year, InsightFace continues to maintain strong and active updates:

Among them, there is a large number of new progress in academic circles, international competitions (ICCV21, NIST-FRVT), and also supports a variety of frameworks (MXNET, Pytorch, PaddlePaddle).

1 InsightFace Framework

Insightface is an open source 2D & 3D deep face analysis toolbox, which highly enables rich and diverse face recognition, face detection and face algorithm, and optimized for training and deployment, assessing, competition in multiple algorithms. Get a winning.

2 Added support based on PaddlePaddle framework implementing a face recognition system

Based on the arser-based Blazeface, ArcFace and MobileFace, unpacking and a certain speed advantage, support large-scale distributed training, high throughput, and support 60 million categories.

(1) ARCFACE model is open and has a certain speed advantage

  • Use the box

  • Speed ??advantage

(2) The RESNET large-scale distributed training rate is high, up to 60 million categories

  • Throughout the lead

  • Maximum support for 60 million categories

(3) a line of code practice simple face recognition Demo

Example scripts for detecting + identifying series predictions are as follows:

Python3.7 Tools / – Det – Rec –index=index.bin –input=Friends2.jpg –output=""

The final visual results are stored in the Output directory, and the visual results are shown below.


3 Welfare time

In order to make everyone more familiar with Insightface, understand the latest update progress, the InsightFac initiates people with the average developers, live in the B station, and share the latest progress for everyone.

4 Registration course joining technology exchange group