BEAT - pkorshunov/simple_antispoofing

This algorithm is a legacy one. The API has changed since its implementation. New versions and forks will need to be updated.

This algorithm is an analyzer. It can only be used on analysis blocks.

Endpoint Groups 2

Algorithms have at least one input and one output. All algorithm endpoints are organized in groups. Groups are used by the platform to indicate which inputs and outputs are synchronized together. The first group is automatically synchronized with the channel defined by the block in which the algorithm is deployed.

Group: test

Endpoint Name	Data Format	Nature
scores_test	system/float/1	Input
label_test	system/text/1	Input

Group: dev

Endpoint Name	Data Format	Nature
scores_dev	system/float/1	Input
label_dev	system/text/1	Input

Results 15

Analyzers may produce any number of results. Once experiments using this analyzer are done, you may display the results or filter experiments using criteria based on them.

Name	Type
test_frr	float32
test_positives	int32
dev_eer_threshold	float32
dev_positives	int32
dev_roc	plot/isoroc/1
test_hter	float32
dev_negatives	int32
test_scores	plot/bar/1
dev_scores	plot/bar/1
dev_far	float32
test_roc	plot/isoroc/1
test_negatives	int32
test_far	float32
dev_frr	float32
dev_eer	float32

Parameters 2

Parameters allow users to change the configuration of an algorithm when scheduling an experiment

Name	Description	Type	Default	Range/Choices
n_bins_negatives	Number of histogram bins for negative scores	uint32	100
n_bins_positives	Number of histogram bins for positive scores	uint32	20

import numpy
import bob.measure

class Algorithm:

def __init__(self):
        self.positives_dev = []
        self.negatives_dev = []
        self.positives_test = []
        self.negatives_test = []
        self.n_bins_positives = 20
        self.n_bins_negatives = 100

def setup(self, parameters):
        self.n_bins_positives = float(parameters.get('n_bins_positives', self.n_bins_positives))
        self.n_bins_negatives = float(parameters.get('n_bins_negatives', self.n_bins_negatives))
        return True

def process(self, inputs, outputs):

# accumulate the dev scores
        while inputs['scores_dev'].hasMoreData():
            inputs['scores_dev'].next()
            inputs['label_dev'].next()
            data_dev = inputs['scores_dev'].data.value
            label_dev = inputs['label_dev'].data.text
            if label_dev == 'real':
                self.positives_dev.append(data_dev)
            else:
                self.negatives_dev.append(data_dev)

# accumulate the test scores
        while inputs['scores_test'].hasMoreData():
            inputs['scores_test'].next()
            inputs['label_test'].next()
            data_test = inputs['scores_test'].data.value
            label_test = inputs['label_test'].data.text
            if label_test == 'real':
                self.positives_test.append(data_test)
            else:  
                self.negatives_test.append(data_test)

# once all values are received, evaluate the scores
        if not(inputs.hasMoreData()):
            eer_threshold      = bob.measure.eer_threshold(self.negatives_dev, self.positives_dev)
            far_dev, frr_dev   = bob.measure.farfrr(self.negatives_dev, self.positives_dev, eer_threshold)
            far_test, frr_test = bob.measure.farfrr(self.negatives_test, self.positives_test, eer_threshold)
            roc_points_dev     = bob.measure.roc(self.negatives_dev, self.positives_dev, 100)
            roc_points_test    = bob.measure.roc(self.negatives_test, self.positives_test, 100)

dev_hist_pos, dev_bin_pos = numpy.histogram(self.positives_dev, self.n_bins_positives)
            dev_hist_neg, dev_bin_neg = numpy.histogram(self.negatives_dev, self.n_bins_negatives)
            
            test_hist_pos, test_bin_pos = numpy.histogram(self.positives_test, self.n_bins_positives)
            test_hist_neg, test_bin_neg = numpy.histogram(self.negatives_test, self.n_bins_negatives)
            
            eer_threshold = bob.measure.eer_threshold(self.negatives_dev, self.positives_dev)
            far_dev, frr_dev   = bob.measure.farfrr(self.negatives_dev, self.positives_dev, eer_threshold)
            far_test, frr_test = bob.measure.farfrr(self.negatives_test, self.positives_test, eer_threshold)
            roc_points_dev     = bob.measure.roc(self.negatives_dev, self.positives_dev, 100)
            roc_points_test    = bob.measure.roc(self.negatives_test, self.positives_test, 100)

# writes the output back to the platform
            outputs.write({
                'dev_eer': numpy.float32((far_dev+frr_dev) / 2.),
                'test_hter': numpy.float32((far_test+frr_test) / 2.),
                'dev_far': numpy.float32(far_dev),
                'dev_frr': numpy.float32(frr_dev),
                'test_far': numpy.float32(far_test),
                'test_frr': numpy.float32(frr_test),
                'dev_positives': numpy.int32(len(self.positives_dev)),
                'dev_negatives': numpy.int32(len(self.negatives_dev)),
                'test_positives': numpy.int32(len(self.positives_test)),
                'test_negatives': numpy.int32(len(self.negatives_test)),
                'dev_eer_threshold': numpy.float32(eer_threshold),
                'dev_roc': {
                        "data":
                        [   
                            {                              
                                "label": "roc_dev",
                                "false_positives": roc_points_dev[1],
                                "false_negatives": roc_points_dev[0],
                                "number_of_positives": numpy.uint64(len(self.positives_dev)),
                                "number_of_negatives": numpy.uint64(len(self.negatives_dev)),
                            }   
                        ]   
                    },
                'test_roc': {
                    "data":
                        [   
                            {   
                                "label": "roc_test",
                                "false_positives": roc_points_test[1],
                                "false_negatives": roc_points_test[0],
                                "number_of_positives": numpy.uint64(len(self.positives_test)),
                                "number_of_negatives": numpy.uint64(len(self.negatives_test)),
                            }   
                        ]   
                    },
                
                'dev_scores': {
                    "data":
                        [
                            {
                                "label": "attacks",
                                "x": dev_bin_neg[:-1],
                                "y": dev_hist_neg,
                            },
                            {
                                "label": "genuines",
                                "x": dev_bin_pos[:-1],
                                "y": dev_hist_pos,
                            }
                        ]
                    },
                'test_scores': {
                    "data":
                        [
                            {
                                "label": "attacks",
                                "x": test_bin_neg[:-1],
                                "y": test_hist_neg,
                            },
                            {
                                "label": "genuines",
                                "x": test_bin_pos[:-1],
                                "y": test_hist_pos,
                            }
                        ]
                    },
            })
        return True

The code for this algorithm in Python
The ruler at 80 columns indicate suggested POSIX line breaks (for readability).
The editor will automatically enlarge to accomodate the entirety of your input
Use keyboard shortcuts for search/replace and faster editing. For example, use Ctrl-F (PC) or Cmd-F (Mac) to search through this box

An algorithm that implements standard metrics for antispoofing evaluation.

Specifically, it outputs, for the development set, the following quantities:

eer: the equal error rate
threshold: the threshold at the equal error rate
dev_far: the false acceptance rate
dev_frr: the false rejection rate
dev_positives: the number of genuine samples
dev_negatives: the number of spoofing attack samples
dev_roc: the receiver operating characteristic curve
dev_scores: the score distribution for all samples on the set

And for the test set:

hter: the half total error rate, calculated on the EER
test_far: the false acceptance rate
test_frr: the false rejection rate
test_positives: the number of genuine samples
test_negatives: the number of spoofing attack samples
test_roc: the receiver operating characteristic curve
test_scores: the score distribution for all samples on the set

This implementation relies on the 'measure' package from the Bob library. In particular, you may check the package bob.measure for more details.

Experiments

Attestation:

Privacy:

Status:

			Updated	Name	Databases/Protocols	Analyzers
				pkorshunov/pkorshunov/speech-antispoofing-baseline/1/btas2016-baseline-pa	avspoof/1@physicalaccess_antispoofing	pkorshunov/simple_antispoofing_analyzer/2

Scientific Python 2.7 (0.1.0) 9

This table shows the number of times this algorithm has been successfully run using the given environment. Note this does not provide sufficient information to evaluate if the algorithm will run when submitted to different conditions.

algorithms pkorshunov simple_antispoofing_analyzer 2