Review — FaceNet: A Unified Embedding for Face Recognition and Clustering

Using Triplet Loss for Contrastive Learning

Triplet Loss (Figure from


1. FaceNet Framework

FaceNet: Framework

2. Triplet Loss

Triplet Loss

2.1. Loss Function

2.2. Triplet Selection

Same and a different person in different pose and illumination combinations (Larger the number, more different the person)

3. Network Architecture

3.1. NN1 (Modified ZFNet)

NN1 (Modified ZFNet)

3.2. NN2, NN3, NN4, NNS1, NNS2 (GoogLeNet / Inception-v1)

NN2 (GoogLeNet / Inception-v1)

4. Experimental Results

4.1. Effect of CNN Model

ROC for the four different models
Mean validation rate VAL at 10E-3 false accept rate (FAR)

4.2. Sensitivity to Image Quality

Effect of Image Quality and Image Size

4.3. Embedding Dimensionality

NN1 on the hold-out set

4.4. Amount of Training Data

4.5. Performance on LFW Dataset

Failure Cases on LFW

4.6. Performance on YouTube Faces DB Dataset

4.7. Face Clustering

One Example of Cluster for Face Clustering

4.8. Harmonic Embedding

Harmonic Embedding Space


Face Recognition

PhD, Researcher. I share what I've learnt and done. :) My LinkedIn:, My Paper Reading List: