Insightface-tf

Intro

• This repositories is a reimplementation of Insightface(github)
• Training code is included, you can use your own data to train models
• We are trying to use new features and Hight Level APIs in TesnsorFlow, such as Keras, Eager Execution, tf.data and so on. The code can even run successful under TF2.0 with a little changes
• The recognition(github-recognition) with tf-1.14 has been reimplemented and the RetinaFace-tf will be upload soon

TODO List

• recognition
  • Backbones
    • ResNet_v1 [done]
  • Losses
    • Arcface loss [done]
    • Cosface loss [done]
    • Sphereface loss [done]
    • Triplet loss [done]
    • Center loss [done]
  • Training code [done]
  • Evaluate [done]
  • Freeze to pb model [todo]
• RetinaFace [todo]

Running Environment

• TensorFlow1.14
• python 3.7
• numpy, pyyaml, matplotlib (Anaconda 3 recommended)