Skip to content

Settings#

System settings#

Parameter Description Type Default value
verboseLogging Level of log verbosity. 1 - Errors, 2 - Warnings, 3 - Info, 4 - Debug. "Value::Int1" 2
betaMode Enable experimental features (0 - Off, 1 - On). "Value::Int1" 0
defaultDetectorType Detector type: FaceDetV1, FaceDetV2, FaceDetV3. "Value::String" FaceDetV1

Verbosity level sets the upper limit of what type of messages may be printed out by the Luna SDK. For example, if user set verboseLogging to 3, it means that Errors, Warnings and Info messages will be printed out to the console. Verbose level of 0 indicates that there are no logging messages printed out at all.

Example:

<section name="system">
    <param name="verboseLogging" type="Value::Int1" x="2" />
    <param name="betaMode" type="Value::Int1" x="0" />
    <param name="detectorType" type="Value::String" text="FaceDetV1" />
</section>

Descriptor factory settings#

Descriptor factory is a facility that creates descriptor extractors and matchers. Both of them utilize algorithms that require a number of coefficients ("weights") to operate properly.

Parameter Description Type Default value
model CNN face descriptor version. "Value::Int1" 62
Possible values: 58, 59, 60, 62
useMobileNet MobileNet is faster but less accurate. Possible values: 0 - don't use mobile net version, 1 - use mobile net version. "Value::Int1" 0
distance Distance between descriptors on matching. L1 faster, L2 make better precision. Possible values: L1, L2. "Value::String" "L2"
descriptorCount WarningLevel Threshold, that limits the ratio of created descriptors to the amount, defined by your license. When the threshold is exceeded, FSDK prints the warning. "Value::Float1" 0.9
calcSimilarity Enable similarity calculation during matching process. Possible values: 1 - enable, 0 - disable. "Value::Int1" 1
calcDistanceSqrt Enable calculation of the square root of distance. Possible values: 1 - enable, 0 - disable "Value::Int1" 1

Models with versions 58, 59, 60, 62 support just L2 distance.

Example:

<section name="DescriptorFactory::Settings">
        <param name="model" type="Value::Int1" x="62" />
        <param name="useMobileNet" type="Value::Int1" x="0" />
        <param name="distance" type="Value::String" text="L2" />
        <param name="descriptorCountWarningLevel" type="Value::Float1" x="0.9" />
        <!--Calculate similarity by default to preserve backward compatible-->
        <param name="calcSimilarity" type="Value::Int1" x="1" />
        <!--Calculate sqrt from distance by default to preserve backward compatible-->
        <param name="calcDistanceSqrt" type="Value::Int1" x="1" />
</section>

FaceDetV3 detector settings#

Parameter Description Type Default value
ScoreThreshold Detection score threshold (RGB) in [0..1] range. "Value::Float1" 0.5
ScoreThresholdNPU Detection score threshold (RGB) in [0..1] range. "Value::Float1" 0.89
ScoreThresholdIR Detection score threshold (InfraRed) in [0..1] range. "Value::Float1" 0.38
RedetectScoreThreshold Redetect score threshold in [0..1] range. "Value::Float1" 0.5
NMSThreshold Overlap threshold for NMS in [0..1] range. "Value::Float1" 0.35
NMSThresholdNPU Overlap threshold for NMS in [0..1] range. "Value::Float1" 0.35
intraNMSThreshold Overlap threshold for intra NMS in [0..1] range. "Value::Float1" 0.35
minFaceSize Minimum face size in pixels. "Value::Int1" 50
nms Type of NMS: mean or best. "Value::String" mean
RedetectTensorSize Target face after preprocessing for redetect. "Value::Int1" 80
Non-public parameter. Do not change.
RedetectFaceTargetSize Target face size for redetect. "Value::Int1" 64
Non-public parameter. Do not change.
paddings Extension of rectangle for RGB mode. Do not change. "Value::Float4" see below
planPrefix Plan prefix. "Value::String" FaceDet_v3_a5
pyramidAlgorithm Turn On/Off the pyramid based algorithm. "Value::Int1" 0
cropPaddingAlignment Non-public parameter. Do not change. "Value::Int1" 64
batchCapacity Non-public parameter. Do not change. "Value::Int1" 16
concurrentBatchSubmission Non-public parameter. Do not change. "Value::Int1" 1
detectMean Non-public parameter. Do not change. "Value::Float3" see below
detectSigma Non-public parameter. Do not change. "Value::Float3" see below
redetectMean Non-public parameter. Do not change. "Value::Float3" see below
redetectSigma Non-public parameter. Do not change. "Value::Float3" see below 
<section name="FaceDetV3::Settings">
    <param name="ScoreThreshold" type="Value::Float1" x="0.5"/>   <!-- used for RGB mode -->
    <param name="ScoreThresholdNPU" type="Value::Float1" x="0.89"/>   <!-- used for RGB mode -->
    <param name="ScoreThresholdIR" type="Value::Float1" x="0.38"/> <!-- used for InfraRed mode -->
    <param name="RedetectScoreThreshold" type="Value::Float1" x="0.5"/>
    <param name="NMSThreshold" type="Value::Float1" x="0.35"/>
    <param name="NMSThresholdNPU" type="Value::Float1" x="0.35"/>
    <param name="intraNMSThreshold" type="Value::Float1" x="0.35"/>
    <param name="minFaceSize" type="Value::Int1" x="50" />
    <param name="pyramidAlgorithm" type="Value::Int1" x="1" />
    <param name="nms" type="Value::String" text="mean"/> <!-- best, mean -->
    <param name="RedetectTensorSize" type="Value::Int1" x="80"/>
    <param name="RedetectFaceTargetSize" type="Value::Int1" x="64"/>
    <param name="paddings" type="Value::Float4" x="-0.18685804" y="0.09821641" z="0.199056897" w="0.07416578" />
    <param name="planPrefix" type="Value::String" text="FaceDet_v3_a5" />
    <param name="cropPaddingAlignment" type="Value::Int1" x="64" />
    <param name="batchCapacity" type="Value::Int1" x="16" />
    <param name="concurrentBatchSubmission" type="Value::Int1" x="1" />
    <param name="detectMean" type="Value::Float3" x="0.0" y="0.0" z="0.0" />
    <param name="detectSigma" type="Value::Float3" x="0.0" y="0.0" z="0.0" />
    <param name="redetectMean" type="Value::Float3" x="0.0" y="0.0" z="0.0" />
    <param name="redetectSigma" type="Value::Float3" x="0.0" y="0.0" z="0.0" />
</section>

FaceDetV3m detector settings#

Parameter Description Type Default value
ScoreThreshold Detection score threshold (RGB) in [0..1] range. "Value::Float1" 0.5
ScoreThresholdNPU Detection score threshold (RGB) in [0..1] range. "Value::Float1" 0.89
ScoreThresholdIR Detection score threshold (InfraRed) in [0..1] range. "Value::Float1" 0.38
RedetectScoreThreshold Redetect score threshold in [0..1] range "Value::Float1" 0.5
NMSThreshold Overlap threshold for NMS in [0..1] range "Value::Float1" 0.35
NMSThresholdNPU Overlap threshold for NMS in [0..1] range "Value::Float1" 0.35
intraNMSThreshold Overlap threshold for intra NMS in [0..1] range "Value::Float1" 0.35
minFaceSize Minimum face size in pixels. "Value::Int1" 50
nms Type of NMS: mean or best "Value::String" mean
RedetectTensorSize Target face after preprocessing for redetect "Value::Int1" 80
Non-public parameter. Do not change.
RedetectFaceTargetSize Target face size for redetect "Value::Int1" 64
Non-public parameter. Do not change.
paddings Extension of rectangle for RGB mode. Do not change. "Value::Float4" see below
planPrefix Plan prefix "Value::String" FaceDet_v3m
pyramidAlgorithm Turn On/Off the pyramid based algorithm "Value::Int1" 1
cropPaddingAlignment Non-public parameter. Do not change. "Value::Int1" 64
batchCapacity Non-public parameter. Do not change. "Value::Int1" 16
concurrentBatchSubmission Non-public parameter. Do not change. "Value::Int1" 1
detectMean Non-public parameter. Do not change. "Value::Float3" see below
detectSigma Non-public parameter. Do not change. "Value::Float3" see below
redetectMean Non-public parameter. Do not change. "Value::Float3" see below
redetectSigma Non-public parameter. Do not change. "Value::Float3" see below 
<section name="FaceDetV3m::Settings">
         <param name="ScoreThreshold" type="Value::Float1" x="0.5"/>   <!-- used for RGB mode -->
         <param name="ScoreThresholdNPU" type="Value::Float1" x="0.89"/>   <!-- used for RGB mode -->
         <param name="ScoreThresholdIR" type="Value::Float1" x="0.38"/> <!-- used for InfraRed mode -->
         <param name="RedetectScoreThreshold" type="Value::Float1" x="0.5"/>
         <param name="NMSThreshold" type="Value::Float1" x="0.35"/>
         <param name="NMSThresholdNPU" type="Value::Float1" x="0.35"/>
         <param name="intraNMSThreshold" type="Value::Float1" x="0.35"/>
         <param name="minFaceSize" type="Value::Int1" x="50" />
         <param name="pyramidAlgorithm" type="Value::Int1" x="1" />
         <param name="nms" type="Value::String" text="mean"/> <!-- best, mean -->
         <param name="RedetectTensorSize" type="Value::Int1" x="80"/>
         <param name="RedetectFaceTargetSize" type="Value::Int1" x="64"/>
         <param name="paddings" type="Value::Float4" x="-0.14193088" y="0.02616714" z="0.15237510" w="0.07140099"/>
         <param name="planPrefix" type="Value::String" text="FaceDet_v3m" />
         <param name="cropPaddingAlignment" type="Value::Int1" x="64" />
         <param name="batchCapacity" type="Value::Int1" x="16" />
         <param name="concurrentBatchSubmission" type="Value::Int1" x="1" />
         <param name="detectMean" type="Value::Float3" x="0.0" y="0.0" z="0.0" />
         <param name="detectSigma" type="Value::Float3" x="0.0" y="0.0" z="0.0" />
         <param name="redetectMean" type="Value::Float3" x="0.0" y="0.0" z="0.0" />
         <param name="redetectSigma" type="Value::Float3" x="0.0" y="0.0" z="0.0" />
</section>

FaceDetV1 detector settings#

Parameter Description Type Default value
FirstThreshold 1-st threshold in [0..1] range. "Value::Float1" 0.6
SecondThreshold 2-nd threshold in [0..1] range. "Value::Float1" 0.7
ThirdThreshold 3-d threshold in [0..1] range. "Value::Float1" 0.93
minFaceSize Minimum face size in pixels. "Value::Int1" 50
scaleFactor Image scale factor. "Value::Float1" 0.7
paddings Extension of rectangle. Do not change. "Value::Float4" see below
redetectTolerance Redetection threshold "Value::Int1" 0
useLNet Whether to use LNet or not. "Value::Int" 1

MinSize and scaleFactor accelerate face detection at the cost of lower recall for smaller faces

Example:

<section name="FaceDetV1::Settings">
    <param name="FirstThreshold" type="Value::Float1" x="0.6"/>
    <param name="SecondThreshold" type="Value::Float1" x="0.7"/>
    <param name="ThirdThreshold" type="Value::Float1" x="0.93"/>
    <param name="minFaceSize" type="Value::Int1" x="50" />
    <param name="scaleFactor" type="Value::Float1" x="0.7" />
    <param name="paddings" type="Value::Float4" x="-0.20099958" y="0.10210337" z="0.20363552" w="0.08490226"/>
    <param name="redetectTolerance" type="Value::Int1" x="0" />
    <param name="useLNet" type="Value::Int1" x="1" />
</section>

FaceDetV2 detector settings#

Parameter Description Type Default value
FirstThreshold 1-st threshold in [0..1] range. "Value::Float1" 0.51385
SecondThreshold 2-nd threshold in [0..1] range. "Value::Float1" 0.248
ThirdThreshold 3-d threshold in [0..1] range. "Value::Float1" 0.76
minFaceSize Minimum face size in pixels. "Value::Int1" 50
scaleFactor Image scale factor. "Value::Float1" 0.7
paddings Extension of rectangle. Do not change. "Value::Float4" see below
redetectTolerance Redetection threshold "Value::Int1" 0
useLNet Whether to use LNet or not. "Value::Int" 1

MinSize and scaleFactor accelerate face detection at the cost of lower recall for smaller faces

Example:

<section name="FaceDetV2::Settings">
        <param name="FirstThreshold" type="Value::Float1" x="0.51385"/>
        <param name="SecondThreshold" type="Value::Float1" x="0.248"/>
        <param name="ThirdThreshold" type="Value::Float1" x="0.76"/>
        <param name="minFaceSize" type="Value::Int1" x="50" />
        <param name="scaleFactor" type="Value::Float1" x="0.7" />
        <param name="paddings" type="Value::Float4" x="-0.20099958" y="0.10210337" z="0.20363552" w="0.08490226" />
        <param name="redetectTolerance" type="Value::Int1" x="0" />
        <param name="useLNet" type="Value::Int1" x="0" />
</section>

LNet#

This group of parameters is non-public. Do not change any of the parameters.

LNetIR#

This group of parameters is non-public. Do not change any of the parameters.

SLNet#

This group of parameters is non-public. Do not change any of the parameters.

IndexBuilder settings#

You can build the HNSW (Hierarchical Navigable Small Worlds) index with descriptor batches and use it for a quick search of the nearest descriptor neighbors.

Parameter Type Default value
numThreads "Value::Int1" 0
construction "Value::Int1" 2000
search "Value::Int1" 6000
searchForEvaluation "Value::Int1" 20

Parameters description:

numThreads - The number of threads to be used for building or searching. If 0 or less, use the std::hardware_concurrency

construction - Internal construction value. The greater it is, the better the graph is, but the slower the construction is. Important: We do not recommend that you change the value, unless you know what you are doing.

search - Internal search value. A greater value means a slower but more complete search. Important: We do not recommend that you change the value, unless you know what you are doing.

searchForEvaluation - Evaluation index length, a greater value means a slower but more precise evaluation.

Example:

<section name="IndexBuilder::Settings">
    <param name="numThreads" type="Value::Int1" x="0" />
    <param name="construction" type="Value::Int1" x="2000" />
    <param name="search" type="Value::Int1" x="6000" />
    <param name="searchForEvaluation" type="Value::Int1" x="20" />
</section>

HumanDetector settings#

Human body detector.

Parameter Type Default value
ScoreThreshold "Value::Float1" x="0.5"
RedetectScoreThreshold "Value::Float1" x="0.12"
NMSThreshold "Value::Float1" x="0.4"
RedetectNMSThreshold "Value::Float1" x="0.4"
imageSize "Value::Int1" x="640"
nms "Value::String" text="best"
RedetectNMS "Value::String" text="mean"
RedetectTensorSize "Value::Int1" x="110"
RedetectHumanTargetSize "Value::Int1" x="85"

Example:

 <section name="HumanDetector::Settings">
    <param name="ScoreThreshold" type="Value::Float1" x="0.5"/>
    <param name="RedetectScoreThreshold" type="Value::Float1" x="0.12"/>
    <param name="NMSThreshold" type="Value::Float1" x="0.4"/>
    <param name="RedetectNMSThreshold" type="Value::Float1" x="0.4"/>
    <param name="imageSize" type="Value::Int1" x="640"/>
    <param name="nms" type="Value::String" text="best"/> <!-- best, mean -->
    <param name="RedetectNMS" type="Value::String" text="mean"/> <!-- best, mean -->
    <param name="RedetectTensorSize" type="Value::Int1" x="110"/>
    <param name="RedetectHumanTargetSize" type="Value::Int1" x="85"/>
</section>

Head detector settings#

Parameter Description Type Default value
ScoreThreshold Detection score threshold (RGB) in the [0..1] range. "Value::Float1" 0.5
NMSThreshold Overlap threshold for NMS in the [0..1] range. "Value::Float1" 0.35
minHeadSize Minimum face size in pixels. "Value::Int1" 60
nms Type of NMS: mean or best. "Value::String" mean
cropPaddingAlignment Non-public parameter. Do not change. "Value::Int1" 64
batchCapacity Non-public parameter. Do not change. "Value::Int1" 16
concurrentBatchSubmission Non-public parameter. Do not change. "Value::Int1" 1 
<section name="HeadDetector::Settings">
        <param name="ScoreThreshold" type="Value::Float1" x="0.5"/>
        <param name="NMSThreshold" type="Value::Float1" x="0.35"/>
        <param name="minHeadSize" type="Value::Int1" x="60" />
        <param name="nms" type="Value::String" text="mean"/> <!-- best, mean -->
        <param name="cropPaddingAlignment" type="Value::Int1" x="64" />
        <param name="batchCapacity" type="Value::Int1" x="16" />
        <param name="concurrentBatchSubmission" type="Value::Int1" x="1" />
</section>

Quality estimator settings#

Quality estimator looks at several image parameters, like lightness (think overexposure), darkness (think underexposure), blurriness, illumination uniformity value, specularity value. Every float value is comparing with according threshold.

Parameter Type Default value
blurThreshold "Value::Float1" x="0.61"
lightThreshold "Value::Float1" x="0.57"
darknessThreshold "Value::Float1" x="0.50"
illuminationThreshold "Value::Float1" x="0.1"
specularityThreshold "Value::Float1" x="0.1"

Example:

<section name="QualityEstimator::Settings">
    <param name="blurThreshold" type="Value::Float1" x="0.61"/>
    <param name="lightThreshold" type="Value::Float1" x="0.57"/>
    <param name="darknessThreshold" type="Value::Float1" x="0.50"/>
    <param name="illuminationThreshold" type="Value::Float1" x="0.1"/>
    <param name="specularityThreshold" type="Value::Float1" x="0.1"/>
</section>

HeadPoseEstimator settings#

HeadPose estimator is able to compute head pose angles in two different ways.

The first one estimates angles by 68-point face-alignment results.

The second one uses raw input image data.

Configuration block listed below allows user to define which method to use. Default configuration settings enables both estimation methods.

Parameter Type Default value
useEstimationByImage "Value::Int1" 1
useEstimationByLandmarks "Value::Int1" 1

Example:

<section name="HeadPoseEstimator::Settings">
    <param name="useEstimationByImage" type="Value::Int1" x="1"/>
    <param name="useEstimationByLandmarks" type="Value::Int1" x="0"/>
</section>

AttributeEstimator settings#

This estimator is able to estimate many person attributes such as:

  • person's age;
  • gender: male, female;

Some of estimator result values depends on threshold values listed below.

Parameter Description Type Default value
genderThreshold gender threshold in [0..1] range. "Value::Float1" 0.5
adultThreshold adult threshold in [0..1] range. "Value::Float1" 0.2

Example:

<section name="AttributeEstimator::Settings">
    <param name="genderThreshold" type="Value::Float1" x="0.5"/>
    <param name="adultThreshold" type="Value::Float1" x="0.2"/>
</section>

EyeEstimator settings#

This estimator aims to determine:

  • Eye state: Open, Closed, Occluded;
  • Precise eye iris location as an array of landmarks;
  • Precise eyelid location as an array of landmarks.

To determine more exact eye state additional auxiliary model eye_status_estimation_flwr*.plan is used. You can enable this auxiliary model through config (faceengine.conf).

Parameter Description Type Default value
useStatusPlan 0 - Off, 1 - On "Value::Int1" 1

Example:

<section name="EyeEstimator::Settings">
    <param name="useStatusPlan" type="Value::Int1" x="1"/>
</section>

GlassesEstimator settings#

Glasses estimator estimates what types of glasses, if any, person is currently wearing. Quality of estimation depends on threshold values listed below. These threshold values set to optimal by default.

Parameter Description Type Default value
noGlassesThreshold noGlasses threshold in [0..1] range. "Value::Float1" 0.986
eyeGlassesThreshold eyeGlasses threshold in [0..1] range. "Value::Float1" 0.57
sunGlassesThreshold sunGlasses threshold in [0..1] range. "Value::Float1" 0.506

Example:

<section name="GlassesEstimator::Settings">
    <param name="noGlassesThreshold" type="Value::Float1" x="0.986"/>
    <param name="eyeGlassesThreshold" type="Value::Float1" x="0.57"/>
    <param name="sunGlassesThreshold" type="Value::Float1" x="0.506"/>
</section>

OverlapEstimator settings#

This estimator tells whether the face is overlapped by any object.

It returns a structure with 2 fields. The first is the value of overlapping in the range from 0.0 (is not overlapped) to 1.0 (maximum, overlapped), the second is a boolean answer.

The boolean answer depends on the threshold listed below. If the value is greater than the threshold, the answer returns true, else false.

Parameter Description Type Default value
overlapThreshold overlap threshold in [0..1] range. "Value::Float1" 0.01

Example:

<section name="OverlapEstimator::Settings">
    <param name="overlapThreshold" type="Value::Float1" x="0.01"/>
</section>

LivenessFPREstimator settings#

Thresholds are listed below.

Parameter Description Type Default value
realThreshold threshold in [0..1] range. "Value::Float1" 0.6 
<section name="LivenessFPREstimator::Settings">
    <param name="realThreshold" type="Value::Float1" x="0.6"/>
</section>

LivenessIREstimator settings#

This estimator determines whether the person's face is real or fake (photo, printed image).

Image must be received from infra-red camera.

The estimator returns a boolean answer (true - is real, false - is fake).

Estimator can be used in "universal" and "ambarella" modes. The mode is chosen depending on the camera type. Thresholds are listed below.

Parameter Description Type Default value
name universal "Value::String" universal
irUniversalThreshold threshold in [0..1] range. "Value::Float1" 0.5328
irAmbarellaThreshold threshold in [0..1] range. "Value::Float1" 0.76 
<section name="LivenessIREstimator::Settings">
    <param name="name" type="Value::String" x="universal"/>
    <param name="irUniversalThreshold" type="Value::Float1" x="0.5328"/>
    <param name="irAmbarellaThreshold" type="Value::Float1" x="0.76"/>
</section>

NIRLivenessEstimator settings#

This estimator determines whether the person's face is real or fake (photo, printed image).

Thr image must be received from the infra-red camera.

The estimator returns a boolean answer (true - is real, false - is fake).

The estimator can be used in two modes. The first mode is fast, while the other is slower but more accurate. By default, recommended to use the more accurate mode, which is 2.

Parameter Description Type Default value
realThreshold threshold in [0..1] range. "Value::Float1" 0.5
defaultEstimatorMode Configuration of plan files usage. Value::Int1 2 
<section name="NIRLivenessEstimator::Settings">
        <param name="realThreshold" type="Value::Float1" x="0.5"/>
        <!-- Currently, available values to select the estimation mode are: 1 and 2. -->
        <param name="defaultEstimatorMode" type="Value::Int1" x="2"/>
</section>

NIRLiveness estimation mode.

Currently, the available values to select the estimation mode are: Default, M1 and M2. The scenario Default means the mode is specified in config file. @see ISettingsProvider.

Implementation description:

NIRLiveness Estimation Mode Enum

enum class NIRLivenessMode {
   Default,  // Specified in config file.
   M1,       // M1.
   M2        // M2.
};

Mouth Estimator settings#

Mouth estimator predicts predominant mouth state.

Estimator accuracy depends on thresholds listed below.

FPR and TPR values are specified for 0.5 threshold

"Thresholds for MouthEstimation"

Parameter Description Type Default value Threshold range TPR FPR
occlusionThreshold Threshold in the [0..1] range "Value::Float1" 0.5 0.4 - 0.6 0.96 0.009
smileThreshold Threshold in the [0..1] range. "Value::Float1" 0.5 0.4 - 0.6 0.97 0.04
openThreshold Threshold in the [0..1] range. "Value::Float1" 0.5 0.4 - 0.6 0.986 0.01

Example:

<section name="MouthEstimator::Settings">
    <param name="occlusionThreshold" type="Value::Float1" x="0.5"/>
    <param name="smileThreshold" type="Value::Float1" x="0.5"/>
    <param name="openThreshold" type="Value::Float1" x="0.5"/>
</section>

DeepFake Estimator settings#

This estimator is designed to predict is the detected face on the input image synthetic or not.

Estimator accuracy depends on realThreshold and defaultEstimatorType listed below.

"DeepFakeEstimator"

Parameter Description Type Default value
realThreshold Threshold in [0..1] range. "Value::Float1" 0.5
defaultEstimatorType Configuration of plan files usage. Value::Int1 2

Example:

<section name="DeepFakeEstimator::Settings">
        <param name="realThreshold" type="Value::Float1" x="0.5"/>
        <param name="defaultEstimatorType" type="Value::Int1" x="2"/>
</section>

DeepFake estimation mode.

Currently, available values for selecting estimation scenario are: Default, M1 and M2. The scenario Default means the mode is specified in config file. @see ISettingsProvider.

Implementation description:

DeepFake Estimation Mode Enum

enum class DeepFakeMode {
   Default,  // Specified in config file.
   M1,       // M1.
   M2        // M2.
};

Medical mask estimator settings#

Medical mask estimator predicts predominant mask features.

Estimator accuracy depends on thresholds listed below.

If accuracy (low FPR) is more important, TPR could be sacrificed by heightening the threshold.

Corresponding FPR and TPR values are also listed below.

Thresholds for MedicalMaskEstimation#

The below parameters specify parameter thresholds in the [0..1] range and are of the "Value::Float1" type.

Parameter Threshold range Default threshold FPR range TPR range
maskThreshold 0.65 - 0.9 0.65 0.014 - 0.01 0.976 - 0.886
noMaskThreshold 0.65 - 0.79 0.65 0.01 - 0.005 0.94 - 0.903
occludedFaceThreshold 0.5 - 0.602 0.5 0.016 - 0.01 0.924 - 0.881

Thresholds for MedicalMaskEstimationExtended#

The below parameters specify parameter thresholds in the [0..1] range and are of the "Value::Float1" type.

Parameter Threshold range Default threshold FPR range TPR range
maskExtendedThreshold 0.65 - 0.784 0.65 0.013 - 0.01 0.923 - 0.894
noMaskExtendedThreshold 0.65 - 0.79 0.65 0.01 - 0.005 0.94 - 0.903
maskNotInPlaceExtendedThreshold 0.65 - 0.85 0.65 0.009 - 0.005 0.918 - 0.833
occludedFaceExtendedThreshold 0.5 - 0.602 0.5 0.016 - 0.01 0.924 - 0.881

Example:

<section name="MedicalMaskEstimatorV3::Settings">
        <param name="maskExtendedThreshold" type="Value::Float1" x="0.65"/>
        <param name="noMaskExtendedThreshold" type="Value::Float1" x="0.65"/>
        <param name="maskNotInPlaceExtendedThreshold" type="Value::Float1" x="0.65"/>
        <param name="occludedFaceExtendedThreshold" type="Value::Float1" x="0.5"/>
        <param name="maskThreshold" type="Value::Float1" x="0.65"/>
        <param name="noMaskThreshold" type="Value::Float1" x="0.65"/>
        <param name="occludedFaceThreshold" type="Value::Float1" x="0.65"/>
</section>

RedEyeEstimator settings#

Red eye estimator evaluates whether person's eyes are red in a photo or not. Red eye estimation depends on threshold value listed below. These threshold value set to optimal by default.

Parameter Description Type Default value
redEyeThreshold redEyeThreshold threshold in [0..1] range. "Value::Float1" 0.5

Example:

<section name="RedEyeEstimator::Settings">
        <param name="redEyeThreshold" type="Value::Float1" x="0.5"/>
</section>

Depth Estimator settings#

Depth estimator performs liveness check via depth image. It exposes different threshold parameters where each one of them let you configure estimator for your specific use case.

Parameter Description Type Default value
maxDepthThreshold maximum depth distance threshold in mm. Should be in [0..inf] range. "Value::Float1" 3000
minDepthThreshold minimum depth distance threshold in mm. Should be in [0..maxDepthThreshold] range. "Value::Float1" 100
zeroDepthThreshold percentage of zero pixels in input image. Threshold in [0..1] range. "Value::Float1" 0.66
confidenceThreshold score threshold above which person is considered to be alive. Threshold in [0..1] range. "Value::Float1" 0.89 
<section name="DepthEstimator::Settings">
    <param name="maxDepthThreshold" type="Value::Float1" x="3000"/>
    <param name="minDepthThreshold" type="Value::Float1" x="100"/>
    <param name="zeroDepthThreshold" type="Value::Float1" x="0.66"/>
    <param name="confidenceThreshold" type="Value::Float1" x="0.89"/>
</section>

LivenessFlyingFaces Estimator settings#

This estimator tells whether the person's face is real or fake (photo, printed image).

It returns a structure with 2 fields.

The first one is the value in the range from 0.0 (is not real) to 1.0 (maximum, real), the second is a boolean answer.

The boolean answer depends on the "realThreshold". If the value is greater than the threshold, the answer returns true, else false.

Parameter Description Type Default value
realThreshold Threshold in the [0..1] range. "Value::Float1" 0.5
aggregationCoeff Coefficient in the [0..1] range. "Value::Float1" 0.7

Example:

<section name="LivenessFlyingFacesEstimator::Settings">
    <param name="realThreshold" type="Value::Float1" x="0.5"/>
    <param name="aggregationCoeff" type="Value::Float1" x="0.7"/>
</section>

LivenessRGBM Estimator settings#

This estimator tells whether the person's face is real or fake (photo, printed image).

It returns a structure with 2 fields.

The first one is the value in the range from 0.0 (is not real) to 1.0 (maximum, real). The second is a boolean answer.

The boolean answer depends on the "threshold". If the value is greater than the threshold, the answer returns true, else false.

This estimator work is based on background accumulation. So the "backgroundCount" parameter is the amount of the frames for the background calculation.

Other parameters are implementation specific, they are not recommended to change.

Parameter Description Type Default value
backgroundCount frames count "Value::Int1" 100
threshold threshold "Value::Float1" 0.8
coeff1 Non-public parameter. Do not change. "Value::Float1" "0.222"
coeff2 Non-public parameter. Do not change. "Value::Float1" "0.222"

Example:

<section name="LivenessRGBMEstimator::Settings">
    <param name="backgroundCount" type="Value::Int1" x="100"/>
    <param name="threshold" type="Value::Float1" x="0.8"/>
    <param name="coeff1" type="Value::Float1" x="0.222"/>
    <param name="coeff2" type="Value::Float1" x="0.222"/>
</section>

LivenessOneShotRGBEstimator settings#

This estimator tells whether the person's face is real or fake (photo, printed image). Thresholds are listed below.

Liveness protects from presentation attacks - when user tries to cheat biometric system by demonstrating fake face to the face capturing camera, but not from image substitution attacks - when fake image is sent directly to the system, bypassing the camera.

LivenessOneShotRGBEstimator supports images, which are captured on Mobile devices or Webcam (PC or laptop). Correct working of the estimator with other source images is not guaranteed.

Supported shooting mode: cooperative, which means that user must interact with the camera and look at it.

User scenarios examples: authentication in mobile application, confirmation of transactions with biometric facial verification.

Image resolution minimum requirements:

  • Mobile devices - 720 × 960 px
  • Webcam (PC or laptop) - 1280 x 720 px
Parameter Description Type Default value
netType NET version. "Value::Int1" 0
realThreshold Threshold in [0..1] range. "Value::Float1" 0.5
qualityThreshold Threshold in [0..1] range. "Value::Float1" 0.5
calibrationCoeff Coefficient in [0..1] range. "Value::Float1" 0.805
mobileRealThreshold Threshold in [0..1] range. "Value::Float1" 0.5
mobileQualityThreshold Threshold in [0..1] range. "Value::Float1" 0.5
mobileCalibrationCoeff Coefficient in [0..1] range. "Value::Float1" 0.98
liteRealThreshold Threshold in [0..1] range. "Value::Float1" 0.5
liteQualityThreshold Threshold in [0..1] range. "Value::Float1" 0.5
liteCalibrationCoeff Coefficient in [0..1] range. "Value::Float1" 0.993 
<section name="LivenessOneShotRGBEstimator::Settings">
    <param name="netType" type="Value::Int1" x="0" />
    <!--Parameters for backend version (netType == 0) -->
    <param name="realThreshold" type="Value::Float1" x="0.5"/>
    <param name="qualityThreshold" type="Value::Float1" x="0.5" />
    <param name="calibrationCoeff" type="Value::Float1" x="0.805"/>
    <!--Parameters for mobile version (netType == 1) -->
    <param name="mobileRealThreshold" type="Value::Float1" x="0.5"/>
    <param name="mobileQualityThreshold" type="Value::Float1" x="0.5" />
    <param name="mobileCalibrationCoeff" type="Value::Float1" x="0.98"/>
    <!--Parameters for lite version (netType == 2) -->
    <param name="liteRealThreshold" type="Value::Float1" x="0.5"/>
    <param name="liteQualityThreshold" type="Value::Float1" x="0.5" />
    <param name="liteCalibrationCoeff" type="Value::Float1" x="0.993"/>
</section>

Credibility Estimator settings#

Credibility estimator is trained to predict reliability of a person. It does so by returning a score value between [0;1] which will be closer to 1 if a person is more likely to be reliable and closer to 0 otherwise. Along with the output score value estimator also returns an enum value, which will give a plain answer if a person is reliable or not for a user convenience. Credibility estimator sets this enum value by comparing an output score with a reliability threshold value listed in faceengine.conf file. User can modify this threshold in CredibilityEstimator::Settings section:

Parameter Description Type Default value
reliableThreshold threshold "Value::Float1" 0.5

Example:

<section name="CredibilityEstimator::Settings">
    <param name="reliableThreshold" type="Value::Float1" x="0.5"/>
</section>

Natural Light Estimator settings#

Natural Light estimator is trained to predict natural of light on the face image.

It does so by returning a score value between [0;1] which will be closer to 1 if a light is more likely to be natural and closer to 0 otherwise.

Along with the output score value estimator also returns an enum value, which will give a plain answer if a person is reliable or not for a user convenience.

NaturalLight estimator sets this enum value by comparing an output score with a reliability threshold value listed in faceengine.conf file. User can modify this threshold in NaturalLightEstimator::Settings section:

Parameter Description Type Default value
naturalLightThreshold threshold "Value::Float1" 0.5

Example:

<section name="NaturalLightEstimator::Settings">
        <param name="naturalLightThreshold" type="Value::Float1" x="0.5"/>
</section>

BlackWhite Estimator settings#

Estimator checks if image is color, grayscale or infrared.

Estimator accuracy depends on thresholds listed below.

Parameter Description Type Default value
colorThreshold threshold in [0..1] range "Value::Float1" 0.5
irThreshold threshold in [0..1] range. "Value::Float1" 0.5

Estimator outputs ImageColorEstimation which consists of 2 scores and color image type as enum with possible values: Color, Grayscale, Infrared.

  • For color image score colorScore will be close to 1.0 and the second one infraredScore - to 0.0;
  • for infrared image score colorScore will be close to 0.0 and the second oneinfraredScore - to 1.0;
  • for grayscale images both of scores will be near 0.0.

So colorThreshold is responsible for separating Color and Grayscale images; irThreshold is responsible for separating Infrared and Grayscale images.

<section name="BlackWhiteEstimator::Settings">
        <param name="colorThreshold" type="Value::Float1" x="0.5"/>
        <param name="irThreshold" type="Value::Float1" x="0.5"/>
</section>

Fish Eye Estimator settings#

Fish Eye estimator is trained to predict fish eye effect on the face image.

It does so by returning a score value between [0;1] which will be closer to 1 if a fisheye effect is more likely to be applied to the image and closer to 0 otherwise.

Along with the output score value estimator also returns an enum value, which will give a plain answer if a person is reliable or not for a user convenience.

Fish Eye estimator sets this enum value by comparing an output score with a reliability threshold value listed in faceengine.conf file. User can modify this threshold in FishEyeEstimator::Settings section:

Parameter Description Type Default value
fishEyeThreshold threshold "Value::Float1" 0.5

Example:

<section name="FishEyeEstimator::Settings">
    <param name="fishEyeThreshold" type="Value::Float1" x="0.5"/>
</section>

Background Estimator settings#

This estimator is designed to evaluate the background in the original image.

Estimator accuracy depends on the thresholds listed below. The scores are defined in [0,1] range. If two scores are above the threshold, then the background is solid, otherwise the background is not solid.

Parameter Description Type Default value
backgroundThreshold threshold in [0..1] range "Value::Float1" 0.5
backgroundColorThreshold threshold in [0..1] range "Value::Float1" 0.3 
<section name="BackgroundEstimator::Settings">
        <param name="backgroundThreshold" type="Value::Float1" x="0.5"/>
        <param name="backgroundColorThreshold" type="Value::Float1" x="0.3"/>
</section>

Human Attribute Estimator settings#

Human Attribute estimator is trained to predict a bunch of human attributes on the human image.

Human Attribute estimator sets outwear color bool values and age by comparing an output score with a corresponding threshold value listed in faceengine.conf file. User can modify this threshold in HumanAttributeEstimator::Settings section:

Parameter Description Type Default value
blackUpperThreshold threshold "Value::Float1" 0.740
blueUpperThreshold threshold "Value::Float1" 0.655
brownUpperThreshold threshold "Value::Float1" 0.985
greenUpperThreshold threshold "Value::Float1" 0.700
greyUpperThreshold threshold "Value::Float1" 0.710
orangeUpperThreshold threshold "Value::Float1" 0.420
purpleUpperThreshold threshold "Value::Float1" 0.650
redUpperThreshold threshold "Value::Float1" 0.600
whiteUpperThreshold threshold "Value::Float1" 0.820
yellowUpperThreshold threshold "Value::Float1" 0.670
blackLowerThreshold threshold "Value::Float1" 0.700
blueLowerThreshold threshold "Value::Float1" 0.840
brownLowerThreshold threshold "Value::Float1" 0.850
greenLowerThreshold threshold "Value::Float1" 0.700
greyLowerThreshold threshold "Value::Float1" 0.690
orangeLowerThreshold threshold "Value::Float1" 0.760
purpleLowerThreshold threshold "Value::Float1" 0.890
redLowerThreshold threshold "Value::Float1" 0.600
whiteLowerThreshold threshold "Value::Float1" 0.540
yellowLowerThreshold threshold "Value::Float1" 0.930
adultThreshold threshold "Value::Float1" 0.940

Example:

<section name="HumanAttributeEstimator::Settings">
                <param name="blackUpperThreshold" type="Value::Float1" x="0.740"/>
                <param name="blueUpperThreshold" type="Value::Float1" x="0.655"/>
                <param name="brownUpperThreshold" type="Value::Float1" x="0.985"/>
                <param name="greenUpperThreshold" type="Value::Float1" x="0.700"/>
                <param name="greyUpperThreshold" type="Value::Float1" x="0.710"/>
                <param name="orangeUpperThreshold" type="Value::Float1" x="0.420"/>
                <param name="purpleUpperThreshold" type="Value::Float1" x="0.650"/>
                <param name="redUpperThreshold" type="Value::Float1" x="0.600"/>
                <param name="whiteUpperThreshold" type="Value::Float1" x="0.820"/>
                <param name="yellowUpperThreshold" type="Value::Float1" x="0.670"/>

                <param name="blackLowerThreshold" type="Value::Float1" x="0.700"/>
                <param name="blueLowerThreshold" type="Value::Float1" x="0.840"/>
                <param name="brownLowerThreshold" type="Value::Float1" x="0.850"/>
                <param name="greenLowerThreshold" type="Value::Float1" x="0.700"/>
                <param name="greyLowerThreshold" type="Value::Float1" x="0.690"/>
                <param name="orangeLowerThreshold" type="Value::Float1" x="0.760"/>
                <param name="purpleLowerThreshold" type="Value::Float1" x="0.890"/>
                <param name="redLowerThreshold" type="Value::Float1" x="0.600"/>
                <param name="whiteLowerThreshold" type="Value::Float1" x="0.540"/>
                <param name="yellowLowerThreshold" type="Value::Float1" x="0.930"/>

                <param name="adultThreshold" type="Value::Float1" x="0.940"/>
        </section>

Portrait Style Estimator settings#

This estimator is designed to evaluate the status of a person's shoulders in the original image.

Estimator accuracy depends on the threshold listed below.

Parameter Description Type Default value
notPortraitStyleThreshold threshold in [0..1] range "Value::Float1" 0.2
portraitStyleThreshold threshold in [0..1] range "Value::Float1" 0.35
hiddenShouldersThreshold threshold in [0..1] range "Value::Float1" 0.2 
<section name="PortraitStyleEstimator::Settings">
        <param name="notPortraitStyleThreshold" type="Value::Float1" x="0.2"/>
        <param name="portraitStyleThreshold" type="Value::Float1" x="0.35"/>
        <param name="hiddenShouldersThreshold" type="Value::Float1" x="0.2"/>
</section>

HumanFace detector settings#

Parameter Description Type Default value
humanThreshold Threshold in the [0..1] range. "Value::Float1" 0.5
nmsHumanThreshold Threshold in the [0..1] range. "Value::Float1" 0.4
faceThreshold Threshold in the [0..1] range. "Value::Float1" 0.5
nmsFaceThreshold Threshold in the [0..1] range. "Value::Float1" 0.3
associationThreshold Threshold in the [0..1] range. "Value::Float1" 0.5
minFaceSize Minimum face size in pixels. "Value::Int1" 50
batchCapacity Non-public parameter. Do not change. "Value::Int1" 8
cropPaddingAlignment Non-public parameter. Do not change. "Value::Int1" 64 
<section name="HumanFaceDetector::Settings">
    <param name="humanThreshold" type="Value::Float1" x="0.5"/>
    <param name="nmsHumanThreshold" type="Value::Float1" x="0.4"/>
    <param name="faceThreshold" type="Value::Float1" x="0.5"/>
    <param name="nmsFaceThreshold" type="Value::Float1" x="0.3"/>
    <param name="associationThreshold" type="Value::Float1" x="0.5"/>
    <param name="minFaceSize" type="Value::Int1" x="50"/>
    <param name="cropPaddingAlignment" type="Value::Int1" x="64" />
    <param name="batchCapacity" type="Value::Int1" x="8" />
</section>

Landmarks detector settings#

Parameter Description Type Default value
useLNet To detect Landmarks68 "Value::Int1" 1
useSLNet To detect Landmarks5 "Value::Int1" 1 
<section name="LandmarksDetector::Settings">
        <param name="useLNet" type="Value::Int1" x="1" />
        <param name="useSLNet" type="Value::Int1" x="1" />
</section>

Note Please pay attention, both parameters cannot be disabled at the same time. In this case, you will receive the error code (fsdk::FSDKError::InvalidConfig), and logs like below:

[30.08.2022 15:47:15] [Error] [FaceLandmarksDetector] Failed to create FaceLandmarksDetector! The both parameters: "useSLNet" and "useLNet" in section "LandmarksDetector::Settings" are disabled at the same time.

Crowd estimator settings#

Parameter Description Type Default value
defaultEstimatorType Type of the estimator "Value::String" TwoNets
minHeadSize Target minHeadSize "Value::Int1" 6
batchCapacity Non-public parameter. Do not change. "Value::Int1" 1
cropPaddingAlignment Non-public parameter. Do not change. "Value::Int1" 0
lowThreshold Non-public parameter. Do not change. "Value::Float1" 0.1
upThreshold Non-public parameter. Do not change. "Value::Float1" 100

There are next possible values for the defaultEstimatorType parameter:

  • Single - working mode with one network usage

  • TwoNets - working mode with two networks usage

<section name="CrowdEstimator::Settings">
        <!-- Available types are: Single, TwoNets -->
        <param name="defaultEstimatorType" type="Value::String" text="TwoNets"/>
        <param name="minHeadSize" type="Value::Int1" x="6"/>
        <param name="batchCapacity" type="Value::Int1" x="1"/>
        <param name="cropPaddingAlignment" type="Value::Int1" x="0"/>
        <param name="lowThreshold" type="Value::Float1" x="0.1"/>
        <param name="upThreshold" type="Value::Float1" x="100"/>
</section>

LivenessDepthRGBEstimator settings#

LivenessDepthRGBEstimator estimator performs liveness check via pair of depth image and RGB image. It exposes different threshold parameters where each one of them let you configure estimator for your specific use case.

Parameter Description Type Default value
maxDepthThreshold maximum depth distance threshold in mm. Should be in [0..inf] range. "Value::Float1" 3000
minDepthThreshold minimum depth distance threshold in mm. Should be in [0..maxDepthThreshold] range. "Value::Float1" 100
zeroDepthThreshold percentage of zero pixels in input image. Threshold in [0..1] range. "Value::Float1" 0.66
confidenceThreshold score threshold above which person is considered to be alive. Threshold in [0..1] range. "Value::Float1" 0.5 
<section name="LivenessDepthRGBEstimator::Settings">
        <param name="maxDepthThreshold" type="Value::Float1" x="3000"/>
        <param name="minDepthThreshold" type="Value::Float1" x="100"/>
        <param name="zeroDepthThreshold" type="Value::Float1" x="0.66"/>
        <param name="confidenceThreshold" type="Value::Float1" x="0.5"/>
</section>

FightsEstimator settings#

FightsEstimator estimator performs a fight detection on the several frame sequences from the target video. It exposes different parameters where each one of them let you configure estimator for your specific use case.

Parameter Description Type Default value
batchSize count of frames in one sequence (batch) "Value::Int1" 5
minBatchCount minimum sequences (batches) count "Value::Int1" 5
cropSize internal crop size. do not change it! "Value::Int1" 224
threshold score threshold above which the video is considered to contain a fight. Threshold in [0..1] range. "Value::Float1" 0.5
scoreNorm normalization parameter. do not change it! "Value::Float1" 1.8 
<section name="FightsEstimator::Settings">
        <param name="batchSize" type="Value::Int1" text="5"/>
        <param name="minBatchCount" type="Value::Int1" x="5"/>
        <param name="cropSize" type="Value::Int1" x="225"/>
        <param name="threshold" type="Value::Float1" x="0.5"/>
        <param name="scoreNorm" type="Value::Float1" x="1.8"/>
</section>