walkoncross / lfw-evaluation-zyf

LFW evaluation

1. extract face features

python requirements: requirements.txt
usage:

python caffe_ftr.py network_def trained_model image_dir image_list_file layer_name save_file

example:

nohup python caffe_ftr.py /opt/caffe/face_example/face_deploy.prototxt /opt/caffe/face_example/face_snapshot_0509_val0.1_batch476/face_train_test_iter_36000.caffemodel /disk2/data/FACE/LFW/lfw-aligned-mtcnn/ /disk2/data/FACE/LFW/lfw-aligned-mtcnn/list.txt fc5 face_snapshot_0509_val0.1_batch476_iter_36000.mat &

2. LFW 10-folds evaluation

python requirements: requirements.txt

cd evaluation_10folds
python validate_on_lfw.py  face_snapshot_0509_val0.1_batch476_iter_36000.mat

3. (optional) accuracy evaluation (Matlab code)

(code from https://github.com/AlfredXiangWu/face_verification_experiment)

In Matlab, run evaluation_matlab/evaluation.m.