Face Recognition Playground

Play with different face recognition techniques and observe the result. example.png

Download Dataset

# Download training dataset
$ mkdir -p download/train
$ wget -O download/train/CASIA-WebFace.tar.gz https://www.dropbox.com/s/6x03igvbvfwx24w/CASIA-WebFace.tar.gz?dl=1

# Download testing dataset
$ mkdir -p download/test
$ wget -O download/test/CFP_FP.tar.gz https://www.dropbox.com/s/e3u7804yk54yqoj/CFP_FP.tar.gz?dl=1
$ wget -O download/test/LFW.tar.gz https://www.dropbox.com/s/d1y5o66dn8vcpvv/LFW.tar.gz?dl=1

# Untar them by yourself

How to run

# Train face recognition model with softmax
$ python3 main.py --config config/example.yml

# Train face recognition model with triplet loss
$ python3 main.py --config config/facenet.yml

Comparision Result

Face Verification Accuracy Pretrained Model
Softmax Loss 0.87 Link
Triplet Loss 0.90 Link

Visualization of Embedding Space

  • Softmax CrossEntrop
    softmax-distribution

  • Triplet Loss
    triplet-distribution