Face representation and similarity

This repo is used to extract feature of face image and compute cosine similarity and l2 distance of two face images. We tried three state-of-the-art methods to extract feature:

• Do We Really Need to Collect Million of Faces for Effective Face Recognition?. We use the output of pool5 as our feature.
• LightCNN
• Facenet

Dependencies

Please install caffe for Do we, and then run:

pip install -r requirements.txt

for other python packages.

Models used for test please download in Facenet and ResNet-101 for Face Recognition. You do not need to put LightCNN model in model/.

Run

• Try LightCNN:

python face_represent_lightcnn.py --img_list=input/list.txt --model="LightCNN-29v2" --num_classes=80013

Model here could also be LightCNN-29 or LightCNN-9.

• Try Facenet:

python face_represent_facenet.py model/20180402-114759.pb

• Try Do we ...:

python face_represent_dowe.py

Results

Facenet is the best! Speed: LightCNN > Dowe > Facenet

TODO

• Crop image with face detector
• Show results