FastDeploy/examples/vision/faceid/insightface

English | 简体中文

InsightFace Ready-to-deploy Model

• InsightFace
  • （1）The *.pt provided by the Official Library can be deployed after the Export ONNX Model；
  • （2）As for InsightFace model trained on customized data, please follow the operations guidelines in Export ONNX Model to complete the deployment.

List of Supported Models

Now FastDeploy supports the deployment of the following models

• ArcFace
• CosFace
• PartialFC
• VPL

Export ONNX Model

Taking ArcFace as an example: Visit ArcFace official github repository, follow the guidelines to download pt model files, and employ torch2onnx.py to get the file in onnx format.

• Download ArcFace model files

  Link: https://pan.baidu.com/share/init?surl=CL-l4zWqsI1oDuEEYVhj-g code: e8pw  
  

• Export files in onnx format

  PYTHONPATH=. python ./torch2onnx.py ms1mv3_arcface_r100_fp16/backbone.pth --output ms1mv3_arcface_r100.onnx --network r100 --simplify 1
  

Download Pre-trained ONNX Model

For developers' testing, models exported by InsightFace are provided below. Developers can download and use them directly. (The accuracy of the models in the table is sourced from the official library) The accuracy metric is sourced from the model description in InsightFace. Refer to the introduction in InsightFace for more details.

Model	Size	Accuracy (AgeDB_30)
CosFace-r18	92MB	97.7
CosFace-r34	131MB	98.3
CosFace-r50	167MB	98.3
CosFace-r100	249MB	98.4
ArcFace-r18	92MB	97.7
ArcFace-r34	131MB	98.1
ArcFace-r50	167MB	-
ArcFace-r100	249MB	98.4
ArcFace-r100_lr0.1	249MB	98.4
PartialFC-r34	167MB	-
PartialFC-r50	249MB	-

Detailed Deployment Tutorials

• Python Deployment
• C++ Deployment

Release Note

• This tutorial and related code are written based on InsightFace CommitID:babb9a5