akuehlka / emvlc-ipad

Ensemble of Multi-View Learning Classifiers for Cross-Domain Iris Presentation Attack Detection

Usage (raw images)

python antispoofing/mcnns/scripts/mcnnsantispoofing.py
--dataset 7
--augmentation 0
--dataset_path $DATASET_PATH
--ground_truth_path $GT_PATH
--iris_location $IRIS_LOCATION
--output_path $OUTPUT_PATH
--n_jobs 6
--classification
--operation segment
--max_axis 260
--bs 32
--epochs $EPOCHS
--lr 0.001
--decay 0.0
--last_layer softmax
--loss_function 2
--optimizer 1
--reg 0.1
--device_number $CUDA_VISIBLE_DEVICES

Usage (bsif images)

python antispoofing/mcnns/scripts/mcnnsantispoofing.py
--dataset 7
--augmentation 0
--dataset_path $DATASET_PATH
--ground_truth_path $GT_PATH
--iris_location $IRIS_LOCATION
--output_path $OUTPUT_PATH
--n_jobs 6
--feature_extraction
--descriptor bsif
--desc_params "[3x3x8]"
--classification
--operation segment
--max_axis 260
--bs 32
--epochs $EPOCHS
--lr 0.001
--decay 0.0
--last_layer softmax
--loss_function 2
--optimizer 1
--reg 0.1
--device_number $CUDA_VISIBLE_DEVICES

Usage (weighted voting)

python antispoofing/mcnns/scripts/mcnnsantispoofing_fusion.py
"${SCRIPT_PATH}/weighted_votingconfig.json"
--augmentation 0
--weighttype acc
--device_number ${CUDA_VISIBLE_DEVICES}

Trained Networks

The trained networks used are available for download here