I have used dlibs face embedding for face recognition as a part of my project. Now, I am looking to write a research paper about my project and I can't seem to find any documentation about dlib library's face embedding model. The only stuff I was able to find is that:
1) It's based on resnet 34
2) The model has high efficiency when distance is .6
and face net triplet loss is different from dlib face embedding and even research papers have no mentions (based o few IEEE papers i looked) about embedding calculations. So, my question is that how 128D embedding values are calculated?