1chimaruGin / FaceNet-and-FaceLoss-collections-tensorflow2.0

Facenet implemented by tensorflow2.0-keras api and in eager-mode support CenterLoss, LSoftmaxLoss, L2SoftmaxLoss, AMSoftmaxLoss, ASoftmaxLoss, ArcFaceSoftmaxLoss

Description

This repository implements facenet using tensorflow2.0-keras api and in eager-mode supporting different backbones and different loss types. The project is still undergo.

How to use

• Install tensorflow2.0-beta1 version.
• Anaconda virtual enviroment is recommended.
• I use tfrecords format data to create input-pipeline.To create tfrecords format training data, run the script datasets.py by:

  bash scripts/tfrcd_data_gen.sh

TODO

• Provide train code in vggface2 datasets using SoftmaxLoss.
• To implement CenterLoss, LSoftmaxLoss, L2SoftmaxLoss, AMSoftmaxLoss, ASoftmaxLoss, ArcFaceSoftmaxLoss.
• Provide test code in LFW datasets.
• Train some models using different backbone and diffrent loss fun in secord.
• Compare and analysis different loss type.
• Refactor and clean the code.