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" |
54 |
| Possible values: 54, 56, 57, 58, 59, 60 | |||
| 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 54, 56, 57, 58, 59, 60 support just L2 distance.
Example:
<section name="DescriptorFactory::Settings">
<param name="model" type="Value::Int1" x="54" />
<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" />
<param name="calcSimilarity" type="Value::Int1" x="1" />
<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" |
0.42 |
| 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" |
0.3 |
| 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 |
| 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 |
| 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="0.42"/> <!-- 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="0.3"/>
<param name="NMSThreshold" type="Value::Float1" x="0.35"/>
<param name="NMSThresholdNPU" type="Value::Float1" x="0.35"/>
<param name="minFaceSize" type="Value::Int1" x="50" />
<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>
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#
HNSW index can be built with descriptors batches and used to search nearest descriptor neighbors very fast.
| Parameter | Description | Type | Default value |
|---|---|---|---|
| numThreads | Number of threads to use on build. If 0 or less, use std::hardware_concurrency value. | "Value::Int1" |
0 |
| construction | Internal construction value. The greater it is, the better is graph, but slower construction. DO NOT CHANGE, unless you know what you are doing. | "Value::Int1" |
2000 |
| search | Internal search value. Greater value means slower but more complete search. DO NOT CHANGE, unless you know what you are doing. | "Value::Int1" |
6000 |
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" />
</section>
HumanDetector settings#
Human body detector.
| Parameter | Type | Default value |
|---|---|---|
| ScoreThreshold | "Value::Float1" |
x="0.45" |
| 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" |
| humanLandmarks17Threshold | "Value::Float1" |
x="0.2" |
| RedetectTensorSize | "Value::Int1" |
x="110" |
| RedetectHumanTargetSize | "Value::Int1" |
x="85" |
Example:
<section name="HumanDetector::Settings">
<param name="ScoreThreshold" type="Value::Float1" x="0.45"/>
<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="humanLandmarks17Threshold" type="Value::Float1" x="0.2"/>
<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 [0..1] range. | "Value::Float1" |
0.5 |
| NMSThreshold | Overlap threshold for NMS in [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>
ChildEstimator settings#
This estimator tells whether the person is child or not.
Child is a person who is yonger than 18 years old.
The estimator returns a structure with 2 fields. The first is the score in the range from 0.0 (is an adult) to 1.0 (maximum, is a child), the second is a boolean answer.
The boolean answer depends on the threshold listed below. If the value is less than the threshold, then true is returned (the person is a child), else false (the person is an adult).
| Parameter | Description | Type | Default value |
|---|---|---|---|
| childThreshold | threshold in [0..1] range. | "Value::Float1" |
0.8508 |
Example:
<section name="ChildEstimator::Settings">
<param name="ChildThreshold" type="Value::Float1" x="0.8508"/>
</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>
HeadAndShouldersLivenessEstimator settings#
This estimator tells whether the person's face is real or fake (photo, printed image). Thresholds are listed below.
| Parameter | Description | Type | Default value |
|---|---|---|---|
| headWidthKoeff | threshold in [0.5..2.0] range | "Value::Float1" |
1.0 |
| headHeightKoeff | threshold in [0.5..2.0] range. | "Value::Float1" |
1.0 |
| shouldersWidthKoeff | threshold in [0.5..2.0] range. | "Value::Float1" |
0.75 |
| shouldersHeightKoeff | threshold in [1.5..5.0] range. | "Value::Float1" |
3.0 |
<section name="HeadAndShouldersLivenessEstimator::Settings">
<param name="headWidthKoeff" type="Value::Float1" x="1.0"/>
<param name="headHeightKoeff" type="Value::Float1" x="1.0"/>
<param name="shouldersWidthKoeff" type="Value::Float1" x="0.75"/>
<param name="shouldersHeightKoeff" type="Value::Float1" x="3.0"/>
</section>
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 [0..1] range | "Value::Float1" |
0.5 |
0.4 - 0.6 |
0.96 |
0.009 |
| smileThreshold | threshold in [0..1] range. | "Value::Float1" |
0.5 |
0.4 - 0.6 |
0.97 |
0.04 |
| openThreshold | threshold in [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>
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 in the table below.
"Thresholds for MedicalMaskEstimation"
| Parameter | Description | Type | Threshold range | Default threshold | FPR range | TPR range |
|---|---|---|---|---|---|---|
| mask | range [0..1] | "Value::Float1" |
0.65 - 0.9 |
0.65 |
0.014 - 0.01 |
0.976 - 0.886 |
| noMask | range [0..1] | "Value::Float1" |
0.65 - 0.79 |
0.65 |
0.01 - 0.005 |
0.94 - 0.903 |
| occludedFace | range [0..1] | "Value::Float1" |
0.5 - 0.602 |
0.5 |
0.016 - 0.01 |
0.924 - 0.881 |
"Thresholds for MedicalMaskEstimationExtended"
| Parameter | Description | Type | Threshold range | Default threshold | FPR range | TPR range |
|---|---|---|---|---|---|---|
| maskExtended | range [0..1] | "Value::Float1" |
0.65 - 0.784 |
0.65 |
0.013 - 0.01 |
0.923 - 0.894 |
| noMaskExtended | range [0..1] | "Value::Float1" |
0.65 - 0.79 |
0.65 |
0.01 - 0.005 |
0.94 - 0.903 |
| maskNotInPlaceExtended | range [0..1] | "Value::Float1" |
0.65 - 0.85 |
0.65 |
0.009 - 0.005 |
0.918 - 0.833 |
| occludedFaceExtended | range [0..1] | "Value::Float1" |
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 [0..1] range. | "Value::Float1" |
0.98 |
| aggregationCoeff | coefficient in [0..1] range. | "Value::Float1" |
0.5 |
Example:
<section name="LivenessFlyingFacesEstimator::Settings">
<param name="realThreshold" type="Value::Float1" x="0.98"/>
<param name="aggregationCoeff" type="Value::Float1" x="0.5"/>
</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 |
|---|---|---|---|
| useMobileNet | use mobile 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" |
0.96 |
| 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" |
0.77 |
<section name="LivenessOneShotRGBEstimator::Settings">
<param name="useMobileNet" type="Value::Int1" x="0" />
<!--Parameters for backend version (useMobileNet == 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="0.96"/>
<!--Parameters for mobile version (useMobileNet == 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="0.77"/>
</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
colorScorewill be close to 1.0 and the second oneinfraredScore- to 0.0; - for infrared image score colorScore will be close to 0.0 and the second one
infraredScore- 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 | Human detection score threshold in [0..1] range. | "Value::Float1" |
0.5 |
| nmsHumanThreshold | Human overlap threshold for NMS in [0..1] range | "Value::Float1" |
0.4 |
| faceThreshold | Face detection score threshold in [0..1] range. | "Value::Float1" |
0.5 |
| nmsFaceThreshold | Face overlap threshold for NMS in [0..1] range | "Value::Float1" |
0.3 |
| associationThreshold | Association score threshold in [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 |
| cropPaddingAlignment | Non-public parameter. Do not change. | "Value::Int1" |
0 |
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">
<param name="defaultEstimatorType" type="Value::String" text="TwoNets"/>
<param name="minHeadSize" type="Value::Int1" x="6"/>
<param name="cropPaddingAlignment" type="Value::Int1" x="0"/>
</section>