Face Embedding is the process of converting a human face image into a vector. Here are the steps to create an Embedding Vector:

1. Face Detector: Use RetinaNet to locate faces on a photo and locate 5 facial landmarks such as nose, eyes, and mouth. We further refined RetinaNet with custom data generated by Data Wow, making the model more accurate in Face finding up to 95%

2. Face Embedding: Convert faces by Face Detector into 128 Dimensions Embedding Vector with Neural Network using Angular Additive Margin Loss function, which helps to isolate more than 50,000 different faces.