By Alexey Gruzdev and Vladislav Sovrasov
A repository for different experimental Face Recognition models such as CosFace, ArcFace, SphereFace, SV-Softmax, etc.
- Create and activate virtual python environment
bash init_venv.sh
. venv/bin/activate
- For Face Recognition training you should download VGGFace2 data. We will refer to this folder as
$VGGFace2_ROOT
. - For Face Recognition evaluation you need to download LFW data and LFW landmarks. Place everything in one folder, which will be
$LFW_ROOT
.
- Go to
$FR_ROOT
folder
cd $FR_ROOT/
- To start training FR model:
python train.py --train_data_root $VGGFace2_ROOT/train/ --train_list $VGGFace2_ROOT/meta/train_list.txt
--train_landmarks $VGGFace2_ROOT/bb_landmark/ --val_data_root $LFW_ROOT/lfw/ --val_list $LFW_ROOT/pairs.txt
--val_landmarks $LFW_ROOT/lfw_landmark.txt --train_batch_size 200 --snap_prefix mobilenet_256 --lr 0.35
--embed_size 256 --model mobilenet --device 1
- To evaluate FR snapshot (let's say we have MobileNet with 256 embedding size trained for 300k):
python evaluate_lfw.py --val_data_root $LFW_ROOT/lfw/ --val_list $LFW_ROOT/pairs.txt
--val_landmarks $LFW_ROOT/lfw_landmark.txt --snap /path/to/snapshot/mobilenet_256_300000.pt --model mobilenet --embed_size 256
Besides passing all the required parameters via command line, the training script allows to read them from a yaml
configuration file.
Each line of such file should contain a valid description of one parameter in the yaml
fromat.
Example:
#optimizer parameters
lr: 0.4
train_batch_size: 256
#loss options
margin_type: cos
s: 30
m: 0.35
#model parameters
model: mobilenet
embed_size: 256
#misc
snap_prefix: MobileFaceNet
devices: [0, 1]
#datasets
train_dataset: vgg
train_data_root: $VGGFace2_ROOT/train/
#... and so on
Path to the config file can be passed to the training script via command line. In case if any other arguments were passed before the config, they will be overwritten.
python train.py -m 0.35 @./my_config.yml #here m can be overwritten with the value from my_config.yml
- You can download pretrained model from fileshare as well - https://download.01.org/openvinotoolkit/open_model_zoo/training_toolbox_pytorch/models/fr/Mobilenet_se_focal_121000.pt
cd $FR_ROOT
python evaluate_lfw.py --val_data_root $LFW_ROOT/lfw/ --val_list $LFW_ROOT/pairs.txt --val_landmarks $LFW_ROOT/lfw_landmark.txt
--snap /path/to/snapshot/Mobilenet_se_focal_121000.pt --model mobilenet --embed_size 256
- You should get the following output:
I1114 09:33:37.846870 10544 evaluate_lfw.py:242] Accuracy/Val_same_accuracy mean: 0.9923
I1114 09:33:37.847019 10544 evaluate_lfw.py:243] Accuracy/Val_diff_accuracy mean: 0.9970
I1114 09:33:37.847069 10544 evaluate_lfw.py:244] Accuracy/Val_accuracy mean: 0.9947
I1114 09:33:37.847179 10544 evaluate_lfw.py:245] Accuracy/Val_accuracy std dev: 0.0035
I1114 09:33:37.847229 10544 evaluate_lfw.py:246] AUC: 0.9995
I1114 09:33:37.847305 10544 evaluate_lfw.py:247] Estimated threshold: 0.7241
- For setting up demo, please go to Face Recognition demo with OpenVINO Toolkit