-
Download and extract dataset (MS1M refined by Insightface, more info here)
-
Prepare dataset
python tools/process_records.py --folder "path to extracted dataset"
-
Update/create configuration file. Set up path to extracted dataset in configuration file (folder: "extracted dataset")
-
Train model
python train.py --config "configuration file"
-
Test model
python test.py --config "configuration file"
Results on trained models:
Models | LFW | CFP_FP | AGEDB_30 |
---|---|---|---|
MobileNet | 0.9958 | 0.9573 | 0.9650 |
Resnet34 | 0.9975 | 0.9764 | 0.9760 |
SEResNeXt50 | 0.9975 | 0.9877 | 0.9792 |