Appendix A. Specifications#
Runtime performance for mobile environment#
Face detection performance depends on input image parameters such as resolution and bit depth as well as the size of the detected face. The Android platform uses mobilenet by default.
Input data characteristics:
- Image resolution: 640x480px;
- Image format: 24 BPP RGB;
Android#
Performance measurements for the ARM of the Samsung Galaxy A52s SM-A528B are presented in the tables below. The measured values are averages of at least 100 experiments. Mobilenet is used by default. The number of threads set to auto means that the maximum number of available threads will be used. For this mode, set the numThreads parameter in runtime.conf to -1. This number of threads equals the number of available processor cores. We strongly recommend that you follow this advice. Otherwise, performance may be significantly reduced. For setting descriptions, see "Configuration Guide - Runtime settings".
Samsung Galaxy A52s SM-A528B, arm64. Matcher performance#
The table below shows the performance of Matcher on Samsung Galaxy A52s SM-A528B.
| Type | Model | CPU threads | Batch Size | Average (matches/sec) |
|---|---|---|---|---|
| Matcher | 59 | 1 | 1 | 1.25M |
| Matcher | 60 | 1 | 1 | 1.67M |
Samsung Galaxy A52s SM-A528B, arm64. Extractor performance#
The table below shows the performance of Extractor on Galaxy A52s SM-A528B.
| Type | CPU threads | Batch Size | Average (ms) | RAM Memory (Mb) |
|---|---|---|---|---|
| 59 | 1 | 1 | 60.17 | 57.0 |
| 59 | auto | 1 | 31.06 | 58.0 |
| 59 | auto | 4 | 25.49 | 67.0 |
| 59 | auto | 8 | 23.38 | 96.0 |
| 60 | 1 | 1 | 54.39 | 53.0 |
| 60 | auto | 1 | 27.76 | 54.0 |
| 60 | auto | 4 | 22.61 | 55.0 |
| 60 | auto | 8 | 20.88 | 75.0 |
Samsung Galaxy A52s SM-A528B, arm64. Detector performance#
The table below shows the performance of Detector on Samsung Galaxy A52s SM-A528B.
| Measurement | Threads | Average (ms) |
|---|---|---|
| Detector (FaceDetV2) | 1 | 10.0 / 36.8 / 38.9 |
| (Easy/complex/6 faces) | auto | 8.95 / 27.5 / 35.4 |
Samsung Galaxy A52s SM-A528B, arm64. Estimations performance#
The table below shows the performance of Estimations on Samsung Galaxy A52s SM-A528B for estimators that have a batch interface.
| Type | CPU threads | Batch Size | Average (ms) | RAM Memory (Mb) |
|---|---|---|---|---|
| HeadPose | 1 | 1 | 0.54 | 44.0 |
| HeadPose | auto | 1 | 1.15 | 43.0 |
| HeadPose | auto | 4 | 0.63 | 61.0 |
| HeadPose | auto | 8 | 0.4 | 85.0 |
| Warper | 1 | 1 | 2.54 | 38.0 |
| Warper | auto | 1 | 3.02 | 38.0 |
| Warper | auto | 4 | 2.9 | 41.0 |
| Warper | auto | 8 | 1.24 | 43.0 |
| Eyes | 1 | 1 | 2.82 | 40.0 |
| Eyes | auto | 1 | 1.74 | 40.0 |
| Eyes | auto | 4 | 1.22 | 39.0 |
| Eyes | auto | 8 | 1.16 | 40.0 |
| Eyes | 1 | 1 | 2.84 | 40.0 |
| Eyes | auto | 1 | 1.59 | 40.0 |
| Eyes | auto | 4 | 1.24 | 39.0 |
| Eyes | auto | 8 | 1.24 | 40.0 |
| AGS | 1 | 1 | 0.56 | 44.0 |
| AGS | auto | 1 | 1.13 | 43.0 |
| AGS | auto | 4 | 0.36 | 61.0 |
| AGS | auto | 8 | 0.31 | 85.0 |
| BestShotQuality | 1 | 1 | 1.46 | 46.0 |
| BestShotQuality | auto | 1 | 1.45 | 45.0 |
| BestShotQuality | auto | 4 | 0.82 | 63.0 |
| BestShotQuality | auto | 8 | 0.5 | 87.0 |
| MedicalMask | 1 | 1 | 16.83 | 64.0 |
| MedicalMask | auto | 1 | 11.47 | 64.0 |
| MedicalMask | auto | 4 | 8.25 | 81.0 |
| MedicalMask | auto | 8 | 7.01 | 105.0 |
| Quality | 1 | 1 | 2.88 | 39.0 |
| Quality | auto | 1 | 3.29 | 39.0 |
| Quality | auto | 4 | 3.46 | 39.0 |
| Quality | auto | 8 | 3.13 | 39.0 |
| Glasses | 1 | 1 | 3.16 | 40.0 |
| Glasses | auto | 1 | 4.15 | 40.0 |
| Glasses | auto | 4 | 3.29 | 40.0 |
| Glasses | auto | 8 | 3.76 | 40.0 |
| LivenessOneShotRGBEstimator | 1 | 1 | 211.37 | 82.0 |
| LivenessOneShotRGBEstimator | auto | 1 | 89.38 | 96.0 |
| LivenessOneShotRGBEstimator | auto | 4 | 84.54 | 212.0 |
| LivenessOneShotRGBEstimator | auto | 8 | 85.35 | 371.0 |
| DepthLivenessEstimator | 1 | 1 | 6.37 | 21.0 |
| DepthLivenessEstimator | auto | 1 | 9.38 | 21.0 |
| DepthLivenessEstimator | auto | 4 | 7.46 | 27.0 |
| DepthLivenessEstimator | auto | 8 | 5.53 | 35.0 |
| Mouth | 1 | 1 | 21.42 | 48.0 |
| Mouth | auto | 1 | 13.06 | 48.0 |
| Mouth | auto | 4 | 9.89 | 54.0 |
| Mouth | auto | 8 | 9.26 | 68.0 |
| FaceOcclusion | 1 | 1 | 21.24 | 40.0 |
| FaceOcclusion | auto | 1 | 13.66 | 40.0 |
| FaceOcclusion | auto | 4 | 11.97 | 71.0 |
| FaceOcclusion | auto | 8 | 12.07 | 108.0 |
Descriptor size#
The table below shows size of serialized descriptors to estimate memory requirements.
"Descriptor size"
| Descriptor version | Data size (bytes) | Metadata size (bytes) | Total size |
|---|---|---|---|
| CNN 62 | 512 | 8 | 520 |
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.
Feature matrix#
Mobile versions come in two editions: the frontend edition (or FE for short) and the complete edition.
The table below shows FaceEngine features supported by different editions of mobile platform.
"Feature matrix"
| Facility | Module | Complete | Frontend |
|---|---|---|---|
| Core | Yes | Yes | |
| Face detection & alignment | Face detector | Yes | Yes |
| Parameter estimation | BestShotQuality estimation | Yes | Yes |
| Color estimation | Yes | Yes | |
| Eye estimation | Yes | Yes | |
| Head pose estimation | Yes | Yes | |
| AGS estimation | Yes | Yes | |
| LivenessOneShotRGB estimation | Yes | Yes | |
| Medical Mask estimation | Yes | Yes | |
| Quality estimation | Yes | Yes | |
| Mouth estimation | Yes | Yes | |
| Glasses estimation | Yes | Yes | |
| Face descriptors | Descriptor extraction | Yes | No |
| Descriptor matching | Yes | No | |
| Descriptor batching | Yes | No | |
| Descriptor search acceleration | Yes | No |
See file "doc/FeatureMapMobile.htm" for more details.