facenet

Why we included this project

FaceNet is the reference TensorFlow implementation of the 2015 FaceNet paper, and it remains a widely studied codebase in face recognition. It turns a face image into a compact 128-dimensional embedding, which is enough to compare faces for verification or to group photos by identity. The repo bundles a complete pipeline: MTCNN-based face detection and alignment plus pretrained Inception-ResNet models, so you can go from raw images to comparable face vectors without wiring the pieces together yourself. The wiki walks through training your own embeddings on a custom dataset and validating accuracy against the LFW benchmark, which makes it a practical way to learn how embedding-based face systems behave in the real world. Just be aware the code targets TensorFlow 1.x and Python 2/3.5, so you will likely need to modernize it before using it in a current stack; as a learning and reference codebase, though, it is remarkably complete.