Appendix A. Specifications#
Classification performance#
Classification performance was measured on a two datasets:
- Cooperative dataset ( containing 20K images from various sources obtained at several banks);
- Non cooperative dataset ( containing 20K ).
The two tables below contain true positive rates corresponding to select false positive rates.
"Classification performance @ low FPR on cooperative dataset"
| FPR | TPR CNN 54 | TPR CNN 56 | TPR CNN 57 | TPR CNN 54m | TPR CNN 56m | TPR CNN 58 | TPR CNN 59 |
|---|---|---|---|---|---|---|---|
| 10^-7^ | 0.9765 | 0.9907 | 0.9906 | 0.9699 | 0.9652 | 0.9910 | 0.9911 |
| 10^-6^ | 0.9849 | 0.9914 | 0.9915 | 0.9829 | 0.9814 | 0.9916 | 0.9915 |
| 10^-5^ | 0.9892 | 0.9916 | 0.9917 | 0.9887 | 0.9886 | 0.9918 | 0.9919 |
| 10^-4^ | 0.9909 | 0.9917 | 0.9918 | 0.9910 | 0.9910 | 0.9919 | 0.9921 |
"Classification performance @ low FPR on non cooperative dataset"
| FPR | TPR CNN 54 | TPR CNN 56 | TPR CNN 57 | TPR CNN 54m | TPR CNN 56m | TPR CNN 58 | TPR CNN 59 |
|---|---|---|---|---|---|---|---|
| 10^-7^ | 0.9638 | 0.9698 | 0.9723 | 0.8813 | 0.8844 | 0.9767 | 0.9832 |
| 10^-6^ | 0.9773 | 0.9809 | 0.9817 | 0.9233 | 0.9229 | 0.9839 | 0.9880 |
| 10^-5^ | 0.9852 | 0.9871 | 0.9873 | 0.9538 | 0.9561 | 0.9880 | 0.9908 |
| 10^-4^ | 0.9896 | 0.9902 | 0.9905 | 0.9752 | 0.9757 | 0.9909 | 0.9924 |
Runtime performance#
Server environment#
Face detection performance depends on input image parameters such as resolution and bit depth as well as the size of the detected face.
Input data characteristics:
- Image resolution: 1200x1600px;
- Image format: 24 BPP RGB;
- Typical face size: ~260x260px.
Performance measurements are presented for CPU, GPU and NPU execution modes in tables below. Measured values are averages of at least 100 experiments.
All batch measurements are performed with minFaceSize = 50.
The results for minimum batch size and optimal batch size are shown in the tables below. All the intermediate and non-optimal values are omitted.
Face detections are performed using FaceDetV3 NN.
CPU performance#
Benchmarking for CPU was performed on the server with the following hardware configuration:
CPU:
- Intel(R) Xeon(R) Silver 4210 CPU @ 2.20GHz;
- CPU(s): 40
- Thread(s) per core: 2
- Core(s) per socket: 10
- Socket(s): 2
- NUMA node(s): 2
- CPU with AVX2 instruction set was used
OS: CentOS Linux release 8.3.2011
RAM: 128 GB DDR4 (Clock Speed: 2133 MHz)
In experiments listed in tables below face detection and descriptor extraction algorithms used all available CPU cores, whereas matching performance is specified per-core.
Descriptor matching is only implemented on CPU.
"CPU mode performance for detection and estimations"
| Measurement | CPU threads | BatchSize | Average (ms) |
|---|---|---|---|
| Detector (minFaceSize=20) | 1 | - | 355.1 |
| Detector (minFaceSize=20) | 8 | - | 154.3 |
| Detector (minFaceSize=50) | 1 | - | 57 |
| Detector (minFaceSize=50) | 8 | - | 24.5 |
| Detector (minFaceSize=90) | 1 | - | 22.1 |
| Detector (minFaceSize=90) | 8 | - | 12 |
| RedetectBatch | 8 | 1 | 5.7 |
| RedetectBatch | 8 | 4 | 6.1 |
| RedetectBatch | 8 | 8 | 3 |
| HumanLandmarksDetector (resize to 640) | 1 | - | 72.6 |
| HumanLandmarksDetector (resize to 640) | 8 | - | 36.2 |
| HumanDetector (resize to 640) | 1 | - | 39.8 |
| HumanDetector (resize to 640) | 8 | - | 18.8 |
| HumanDetector Batch (resize to 640) | 8 | 1 | 20.8 |
| 8 | 4 | 20.5 | |
| 8 | 8 | 20.4 | |
| HumanDetector redetect Batch | 8 | 1 | 1.3 |
| 8 | 4 | 1.3 | |
| 8 | 8 | 1.1 | |
| HumanLandmarksDetector (resize to 320) | 1 | - | 44.5 |
| HumanLandmarksDetector (resize to 320) | 8 | - | 25.1 |
| HumanDetector (resize to 320) | 1 | - | 6.5 |
| HumanDetector (resize to 320) | 8 | - | 6.5 |
| HumanDetector Batch (resize to 320) | 8 | 1 | 6.2 |
| 8 | 4 | 6.0 | |
| 8 | 8 | 5.9 | |
| HeadPoseByLandmarks | 1 | - | 1.57 |
| HeadPoseByLandmarks | 8 | - | 1.59 |
| EyesGaze | 1 | - | 2.5 |
| EyesGaze | 8 | - | 1.7 |
| Emotions | 1 | - | 13.8 |
| Emotions | 8 | - | 5.9 |
| Attributes | 1 | - | 64 |
| Attributes | 8 | - | 27.4 |
| Quality | 1 | - | 1.6 |
| Quality | 8 | - | 0.7 |
| HeadPoseByImage | 1 | - | 0.48 |
| HeadPoseByImage | 8 | - | 0.31 |
| HeadPoseBatch | 8 | 1 | 0.3 |
| HeadPoseBatch | 8 | 8 | 0.09 |
| Warper | 1 | - | 1.7 |
| Warper | 8 | - | 2.1 |
| Eyes | 1 | - | 0.8 |
| Eyes | 8 | - | 0.71 |
| EyesBatch | 8 | 1 | 0.6 |
| EyesBatch | 8 | 8 | 0.3 |
| Infra-Red | 1 | - | 2 |
| Infra-Red | 8 | - | 1 |
| Infra-RedBatch | 8 | 1 | 1.1 |
| Infra-RedBatch | 8 | 8 | 0.8 |
| AGS | 1 | - | 0.25 |
| AGS | 8 | - | 0.2 |
| AGSBatch | 8 | 1 | 0.21 |
| AGSBatch | 8 | 8 | 0.06 |
| Overlap | 1 | - | 4.5 |
| Overlap | 8 | - | 1.2 |
| Glasses | 1 | - | 1.7 |
| Glasses | 8 | - | 1 |
| Eyes | 1 | - | 1.7 |
| Eyes | 8 | - | 1.2 |
| Child | 1 | - | 20.2 |
| Child | 8 | - | 11.7 |
| ChildBatch | 8 | 1 | 11.8 |
| ChildBatch | 8 | 8 | 6.6 |
| BestShotQuality | 1 | - | 0.35 |
| BestShotQuality | 8 | - | 0.22 |
| BestShotQualityBatch | 8 | 1 | 0.24 |
| BestShotQualityBatch | 8 | 8 | 0.07 |
| Mouth | 1 | - | 1.7 |
| Mouth | 8 | - | 0.9 |
| LivenessFlyingFaces | 1 | - | 9.8 |
| LivenessFlyingFaces | 8 | - | 5.0 |
| LivenessRGBMEstimator | 1 | - | 20 |
| LivenessRGBMEstimator | 8 | - | 13.2 |
| MedicalMask | 1 | - | 1.1 |
| MedicalMask | 8 | - | 0.4 |
| MedicalMaskBatch | 8 | 1 | 0.52 |
| MedicalMaskBatch | 8 | 8 | 0.38 |
| LivenessOneShotRGBEstimator | 1 | - | 187.3 |
| LivenessOneShotRGBEstimator | 8 | - | 87.9 |
| LivenessOneShotRGBEstimatorBatch | 8 | 1 | 90.3 |
| LivenessOneShotRGBEstimatorBatch | 8 | 8 | 85.6 |
| Orientation | 1 | - | 19.6 |
| Orientation | 8 | - | 11.3 |
| OrientationBatch | 8 | 1 | 10.9 |
| OrientationBatch | 8 | 8 | 8.1 |
| CredibilityCheck | 1 | - | 111.0 |
| CredibilityCheck | 8 | - | 39.7 |
| CredibilityCheckBatch | 8 | 1 | 39.6 |
| CredibilityCheckBatch | 8 | 8 | 34.1 |
| FacialHair | 1 | - | 2.9 |
| FacialHair | 8 | - | 2.1 |
| FacialHairBatch | 8 | 1 | 2.1 |
| FacialHairBatch | 8 | 8 | 1.0 |
"Extractor performance"
| Type | Model | CPU threads | Average (ms) |
|---|---|---|---|
| Extractor | 57 | 1 | 217 |
| Extractor | 57 | 8 | 79.9 |
| Extractor | 58 | 1 | 214.7 |
| Extractor | 58 | 8 | 80.3 |
| Extractor | 59 | 1 | 213.6 |
| Extractor | 59 | 8 | 68.3 |
| Extractor | 102 | 1 | 3.0 |
| Extractor | 102 | 8 | 3.0 |
| Extractor | 103 | 1 | 140.0 |
| Extractor | 103 | 8 | 50.0 |
| Extractor | 104 | 1 | 14.4 |
| Extractor | 104 | 8 | 6.5 |
The following table includes average matcher per second for descriptors received using the following CNN model versions:
- face descriptors: 57, 58, 59
- human body descriptors: 102, 103, 104
"Matcher performance"
| Type | Model | CPU threads | Batch Size | Average (matches/sec) |
|---|---|---|---|---|
| Matcher | 57, 58, 59 | 1 | 1000 | 42.2 M |
| Matcher | 57, 58, 59 | 1 | 100 000 | 26.60 M |
| Matcher | 102, 103, 104 | 1 | 1000 | 10.17 M |
| Matcher | 102, 103, 104 | 1 | 100 000 | 5.48 M |
GPU performance#
Benchmarking for GPU was performed on the following hardware configuration:
GPU: NVIDIA Tesla T4.
OS: CentOS Linux release 8.3.2011
"GPU mode performance for detection and estimations"
| Measurement | Batch Size | Average (ms) |
|---|---|---|
| Detector (minFaceSize=90) | - | 13.6 |
| Detector (minFaceSize=50) | - | 8.7 |
| Detector (minFaceSize=20) | - | 62.3 |
| DetectorBatch | 1 | 10.1 |
| DetectorBatch | 8 | 7.5 |
| RedetectBatch | 1 | 3.9 |
| RedetectBatch | 32 | 0.2 |
| HumanLandmarksDetector (resize to 640) | - | 35.7 |
| HumanDetector (resize to 640) | - | 20.2 |
| HumanLandmarksDetector (resize to 320) | - | 24.7 |
| HumanDetector (resize to 320) | - | 9.3 |
| Human redetection | - | 1.3 |
| HeadPoseByLandmarks | - | 1.5 |
| EyesGaze | - | 1.7 |
| Emotions | - | 1.9 |
| Attributes | - | 3.4 |
| Quality | - | 0.9 |
| HeadPoseByImage | - | 1.8 |
| HeadPoseBatch | 1 | 1.84 |
| HeadPoseBatch | 32 | 1.05 |
| Warper | - | 1.9 |
| Eyes | - | 0.8 |
| EyesBatch | 1 | 0.78 |
| EyesBatch | 16 | 0.2 |
| Infra-Red | - | 1.2 |
| Infra-RedBatch | 1 | 1.14 |
| Infra-RedBatch | 32 | 0.53 |
| AGS | - | 1.76 |
| AGSBatch | 1 | 1.76 |
| AGSBatch | 16 | 1.05 |
| Overlap | - | 1.08 |
| Glasses | - | 1.05 |
| Eyes | - | 1.39 |
| Child | - | 2.31 |
| ChildBatch | 1 | 2.36 |
| ChildBatch | 16 | 1.37 |
| BestShotQuality | - | 1.86 |
| BestShotQualityBatch | 1 | 2.13 |
| BestShotQualityBatch | 16 | 1.1 |
| Mouth | - | 1.05 |
| LivenessFlyingFaces | - | 4.47 |
| LivenessRGBMEstimator | - | 8.96 |
| MedicalMask | - | 0.84 |
| MedicalMaskBatch | 1 | 0.8 |
| MedicalMaskBatch | 16 | 0.2 |
| LivenessOneShotRGBEstimator | - | 43.2 |
| LivenessOneShotRGBEstimatorBatch | 1 | 42.1 |
| Orientation | - | 5.56 |
| OrientationBatch | 1 | 5.6 |
| OrientationBatch | 16 | 3.71 |
| CredibilityCheck | - | 6.1 |
| CredibilityCheckBatch | 1 | 6.1 |
| CredibilityCheckBatch | 16 | 4.2 |
| FacialHair | - | 1.7 |
| FacialHairBatch | 1 | 1.7 |
| FacialHairBatch | 16 | 0.34 |
"Extractor performance"
| Type | Model | Batch Size | Average (ms) |
|---|---|---|---|
| Extractor | 57 | - | 10.99 |
| ExtractorBatch | 57 | 1 | 11.06 |
| ExtractorBatch | 57 | 16 | 8.34 |
| Extractor | 58 | - | 11.1 |
| ExtractorBatch | 58 | 1 | 11.1 |
| ExtractorBatch | 58 | 16 | 8.35 |
| Extractor | 59 | - | 11.1 |
| ExtractorBatch | 59 | 1 | 11.1 |
| ExtractorBatch | 59 | 16 | 11.4 |
| Extractor | 102 | - | 3.5 |
| ExtractorBatch | 102 | 1 | 3.5 |
| ExtractorBatch | 102 | 16 | 1.1 |
| Extractor | 103 | - | 7.0 |
| ExtractorBatch | 103 | 1 | 7.0 |
| ExtractorBatch | 103 | 16 | 4.6 |
| Extractor | 104 | - | 3.2 |
| ExtractorBatch | 104 | 1 | 3.0 |
| ExtractorBatch | 104 | 16 | 1.2 |
Embedded environment#
Face detection performance depends on input image parameters such as resolution and bit depth as well as the size of the detected face.
Input data characteristics:
- Image resolution: 640x480px;
- Image format: 24 BPP RGB;
- Typical face size: ~260x260px.
All batch measurements are performed with minFaceSize = 50.
The results for minimum batch size and optimal batch size are shown in the tables below. All the intermediate and non-optimal values are omitted.
Face detections are performed using FaceDetV3 NN.
Jetson#
Jetson does not use mobilenet by default.
Performance measurements are presented for Jetson. Measured values are averages of at least 100 experiments. Mobilenet is not used by default.
Jetson TX#
"Jetson TX GPU Performance. Detection and estimation"
| Type | Batch Size | Average (ms) |
|---|---|---|
| Detector (minFaceSize=90) | - | 45.197 |
| Detector (minFaceSize=50) | - | 105.26 |
| Detector (minFaceSize=20) | - | 613.528 |
| Redetect batch | 1 | 18.3 |
| Redetect batch | 32 | 6.08 |
| HumanLandmarksDetector (resize to 640) | - | 222.4 |
| HumanDetector (resize to 640) | - | 66.9 |
| HumanLandmarksDetector (resize to 320) | - | 170.7 |
| HumanDetector (resize to 320) | - | 23.5 |
| HeadPoseByLandmarks | - | 3.32 |
| EyesGaze | - | 6 |
| Emotions | - | 18.14 |
| Attributes | - | 41.7 |
| Quality | - | 3.93 |
| HeadPoseByImage | - | 5.35 |
| HeadPoseBatch | 1 | 5.58 |
| HeadPoseBatch | 32 | 2.88 |
| Warper | - | 4.41 |
| Eyes | - | 3.83 |
| EyesBatch | 1 | 3.96 |
| EyesBatch | 32 | 1.77 |
| Infra-Red | - | 4.91 |
| Infra-RedBatch | 1 | 4.75 |
| AGS | - | 4.49 |
| AGSBatch | 1 | 4.6 |
| AGSBatch | 16 | 2.79 |
| Overlap | - | 4.39 |
| Glasses | - | 6 |
| Eyes | - | 9 |
| EyesBatch | 1 | 9.31 |
| EyesBatch | 16 | 3.56 |
| Child | - | 21.50 |
| ChildBatch | 1 | 21.51 |
| ChildBatch | 16 | 17.78 |
| BestShotQuality | - | 5.76 |
| BestShotQualityBatch | 1 | 5.78 |
| BestShotQualityBatch | 16 | 2.86 |
| Mouth | - | 4.98 |
| LivenessFlyingFaces | - | 22.94 |
| LivenessRGBMEstimator | - | 76.65 |
| MedicalMask | - | 2.25 |
| MedicalMaskBatch | 1 | 2 |
| MedicalMaskBatch | 32 | 0.96 |
| LivenessOneShotRGBEstimator | - | 250.0 |
| LivenessOneShotRGBEstimatorBatch | 1 | 248.3 |
| Orientation | - | 34.9 |
| OrientationBatch | 1 | 34.9 |
| CredibilityCheck | - | 61.0 |
| CredibilityCheckBatch | 1 | 61.2 |
| CredibilityCheckBatch | 8 | 67.4 |
| CredibilityCheckBatch | 16 | 62.8 |
| CredibilityCheckBatch | 32 | 61.9 |
| FacialHair | - | 6.7 |
| FacialHairBatch | 1 | 6.6 |
| FacialHairBatch | 16 | 9.8 |
"Jetson TX GPU Performance. Extractor"
| Type | Model | Batch Size | Average (ms) |
|---|---|---|---|
| Extractor | 57 | - | 133 |
| Extractor Batch | 57 | 1 | 132.5 |
| 57 | 8 | 82.41 | |
| Extractor | 58 | - | 132.4 |
| Extractor Batch | 58 | 1 | 131.6 |
| 58 | 8 | 82.07 | |
| Extractor | 59 | - | 119.5 |
| Extractor Batch | 59 | 1 | 119.6 |
| 59 | 8 | 94.6 | |
| Extractor | 102 | - | 11.8 |
| Extractor Batch | 102 | 1 | 11.8 |
| 102 | 8 | 2.9 | |
| Extractor | 103 | - | 54.0 |
| Extractor Batch | 103 | 1 | 54.0 |
| 103 | 8 | 44.1 | |
| Extractor | 104 | - | 16.4 |
| Extractor Batch | 104 | 1 | 16.4 |
| 104 | 8 | 7.9 |
Jetson Xavier#
"Jetson Xavier GPU Performance. Detection and estimation"
| Type | Batch Size | Average (ms) |
|---|---|---|
| Detector (minFaceSize=90) | - | 16.72 |
| Detector (minFaceSize=50) | - | 27.4 |
| Detector (minFaceSize=20) | - | 145.02 |
| DetectorBatch | 1 | 29 |
| DetectorBatch | 8 | 29.97 |
| RedetectBatch | 1 | 8.29 |
| RedetectBatch | 32 | 0.76 |
| HumanLandmarksDetector (resize to 640) | - | 36.81 |
| HumanDetector (resize to 640) | - | 14.51 |
| HumanLandmarksDetector (resize to 320 ) | - | 39.26 |
| HumanDetector (resize to 320 ) | - | 8.26 |
| HeadPoseByLandmarks | - | 2.63 |
| EyesGaze | - | 5.99 |
| Emotions | - | 5.18 |
| Attributes | - | 10.85 |
| Quality | - | 5.52 |
| HeadPoseByImage | - | 6.69 |
| HeadPoseBatch | 1 | 4.36 |
| HeadPoseBatch | 32 | 0.9 |
| Warper | - | 3.56 |
| Eyes | - | 1.28 |
| EyesBatch | 1 | 1.6 |
| EyesBatch | 32 | 0.6 |
| Infra-Red | - | 3.37 |
| Infra-RedBatch | 1 | 3 |
| Infra-RedBatch | 32 | 1.54 |
| AGS | - | 4.15 |
| AGSBatch | 1 | 3.47 |
| AGSBatch | 32 | 0.92 |
| Overlap | - | 3.29 |
| Glasses | - | 2.26 |
| Eyes | - | 2 |
| EyesBatch | 1 | 1.83 |
| EyesBatch | 32 | 0.98 |
| Child | - | 5.77 |
| ChildBatch | 1 | 5.49 |
| ChildBatch | 8 | 3.91 |
| BestShotQuality | - | 7.26 |
| BestShotQualityBatch | 1 | 7.46 |
| BestShotQualityBatch | 32 | 0.88 |
| Mouth | - | 2.26 |
| LivenessFlyingFaces | - | 8.12 |
| LivenessRGBMEstimator | - | 17.78 |
| MedicalMask | - | 0.87 |
| MedicalMaskBatch | 1 | 1.02 |
| MedicalMaskBatch | 32 | 0.39 |
| LivenessOneShotRGBEstimator | - | 77.04 |
| LivenessOneShotRGBEstimatorBatch | 1 | 77 |
| LivenessOneShotRGBEstimatorBatch | 8 | 76.47 |
| Orientation | - | 8.71 |
| OrientationBatch | 1 | 8.69 |
| OrientationBatch | 32 | 7.2 |
| CredibilityCheck | - | 18.80 |
| CredibilityCheckBatch | 1 | 18.85 |
| CredibilityCheckBatch | 8 | 21.85 |
| CredibilityCheckBatch | 16 | 21.55 |
| CredibilityCheckBatch | 32 | 20.15 |
| FacialHair | - | 3.2 |
| FacialHairBatch | 1 | 3.2 |
| FacialHairBatch | 16 | 0.9 |
"Jetson Xavier GPU Performance. Extractor"
| Type | Model | Batch Size | Average (ms) |
|---|---|---|---|
| Extractor | 57 | - | 36.99 |
| Extractor Batch | 57 | 1 | 36.46 |
| Extractor Batch | 57 | 4 | 32.58 |
| Extractor | 58 | - | 39.67 |
| Extractor Batch | 58 | 1 | 38.36 |
| Extractor Batch | 58 | 8 | 32.26 |
| Extractor | 59 | - | 36.31 |
| Extractor Batch | 59 | 1 | 35.48 |
| Extractor Batch | 59 | 8 | 33.86 |
| Extractor | 102 | - | 6.7 |
| Extractor Batch | 102 | 1 | 6.7 |
| 102 | 8 | 1.2 | |
| Extractor | 103 | - | 17.4 |
| Extractor Batch | 103 | 1 | 17.4 |
| 103 | 8 | 13.7 | |
| Extractor | 104 | - | 9.6 |
| Extractor Batch | 104 | 1 | 9.6 |
| 104 | 8 | 3.4 |
Jetson Xavier NX#
"Jetson Xavier NX GPU Performance. Detection and estimation"
| Type | Batch Size | Average (ms) |
|---|---|---|
| Detector (minFaceSize=90) | - | 16.1 |
| Detector (minFaceSize=50) | - | 39.9 |
| Detector (minFaceSize=20) | - | 224.6 |
| DetectorBatch | 1 | 42.54 |
| DetectorBatch | 8 | 37.75 |
| RedetectBatch | 1 | 7.15 |
| RedetectBatch | 32 | 1.32 |
| HumanLandmarksDetector (resize to 640) | - | 59.2 |
| HumanDetector (resize to 640) | - | 26.3 |
| HumanLandmarksDetector (resize to 320) | - | 38.5 |
| HumanDetector (resize to 320) | - | 10.3 |
| HeadPoseByLandmarks | - | 3.55 |
| EyesGaze | - | 3.3 |
| Emotions | - | 7.83 |
| Attributes | - | 21.35 |
| Quality | - | 2.2 |
| HeadPoseByImage | - | 3.1 |
| HeadPoseBatch | 1 | 3.03 |
| HeadPoseBatch | 32 | 1.25 |
| Warper | - | 5.11 |
| Eyes | - | 1.53 |
| EyesBatch | 1 | 1.52 |
| EyesBatch | 32 | 0.5 |
| Infra-Red | - | 2.86 |
| Infra-RedBatch | 1 | 2.85 |
| Infra-RedBatch | 32 | 1.8 |
| AGS | - | 2.7 |
| AGSBatch | 1 | 2.77 |
| AGSBatch | 32 | 1.22 |
| Overlap | - | 2.7 |
| Glasses | - | 2.75 |
| Eyes | - | 2.5 |
| Eyes | 1 | 2.63 |
| Eyes | 32 | 1.04 |
| Child | - | 9.49 |
| ChildBatch | 1 | 9.49 |
| ChildBatch | 8 | 7.19 |
| BestShotQuality | - | 3.78 |
| BestShotQualityBatch | 1 | 3.7 |
| BestShotQualityBatch | 32 | 1.27 |
| Mouth | - | 1.94 |
| LivenessFlyingFaces | - | 10.9 |
| LivenessRGBMEstimator | - | 29.9 |
| MedicalMask | - | 1.45 |
| MedicalMaskBatch | 1 | 1.46 |
| MedicalMaskBatch | 32 | 0.53 |
| LivenessOneShotRGBEstimator | - | 126.6 |
| LivenessOneShotRGBEstimatorBatch | 1 | 123.4 |
| Orientation | - | 14.89 |
| OrientationBatch | 1 | 14.96 |
| OrientationBatch | 32 | 12.97 |
| CredibilityCheck | - | 40.70 |
| CredibilityCheckBatch | 1 | 40.75 |
| CredibilityCheckBatch | 8 | 50.64 |
| CredibilityCheckBatch | 16 | 49.76 |
| CredibilityCheckBatch | 32 | 46.55 |
| FacialHair | - | 3.3 |
| FacialHairBatch | 1 | 3.3 |
| FacialHairBatch | 16 | 1.9 |
"Jetson Xavier NX GPU Performance. Extractor"
| Type | Model | Batch Size | Average (ms) |
|---|---|---|---|
| Extractor | 57 | - | 78.6 |
| Extractor Batch | 57 | 1 | 78.2 |
| Extractor Batch | 57 | 16 | 68.2 |
| Extractor | 58 | - | 78.3 |
| Extractor Batch | 58 | 1 | 78.1 |
| Extractor Batch | 58 | 16 | 67.7 |
| Extractor | 59 | - | 77.7 |
| Extractor Batch | 59 | 1 | 77.7 |
| Extractor Batch | 59 | 16 | 78.2 |
| Extractor | 102 | - | 6.6 |
| Extractor Batch | 102 | 1 | 6.6 |
| 102 | 16 | 2.0 | |
| Extractor | 103 | - | 34.2 |
| Extractor Batch | 103 | 1 | 34.2 |
| 103 | 16 | 29.4 | |
| Extractor | 104 | - | 9.3 |
| Extractor Batch | 104 | 1 | 9.3 |
| 104 | 16 | 6.8 |
Descriptor size#
Table \ref{Tab.A.3.1} shows size of serialized face descriptors to estimate memory requirements.
"Descriptor size" \label{Tab.A.3.1}
| Face descriptor version | Data size (bytes) | Metadata size (bytes) | Total size |
|---|---|---|---|
| CNN 54 | 512 | 8 | 520 |
| CNN 56 | 512 | 8 | 520 |
| CNN 57 | 512 | 8 | 520 |
| CNN 58 | 512 | 8 | 520 |
| CNN 59 | 512 | 8 | 520 |
Table \ref{Tab.A.3.2} shows size of serialized human descriptors to estimate memory requirements. Human descriptors are used only for reidentification tasks/
"Human descriptor size (used only for reidentification tasks)" \label{Tab.A.3.2}
| Human descriptor version | Data size (bytes) | Metadata size (bytes) | Total size |
|---|---|---|---|
| CNN 102 | 2048 | 8 | 2056 |
| CNN 103 | 2048 | 8 | 2056 |
| CNN 104 | 2048 | 8 | 2056 |
Metadata includes signature and version information that may be omitted during serialization if the NoSignature flag is specified.
When estimating individual descriptor size in memory or serialization storage requirements with default options, consider using values from the "Total size" column.
When estimating memory requirements for descriptor batches, use values from the "Data size" column instead, since a descriptor batch does not duplicate metadata per descriptor and thus is more memory-efficient.
These numbers are for approximate computation only, since they do not include overhead like memory alignment for accelerated SIMD processing and the like.