FaceMe® Accuracy from NIST
National Institute of Standards and Technology (NIST)
True Acceptance Rate
Higher is Better
False Acceptance Rate
Lower is Better
1:N VISA Border (1.6M DB)
* VISA images are photos taken from the passport. The mean interocular distance (IOD) in the VISA images are 69 pixels.
* Wild images are non-constraint, photojournalism-style photos. Resolution varies widely. A wide yaw and pitch angle may be present.
* VISA Border are comparing VISA image to WebCam images.
FaceMe® Accuracy
Mobile, Light-weight IoT/AIoT Devices
True Acceptance Rate
(@FAR 1E-4)
True Acceptance Rate
(@FAR 1E-5)
True Acceptance Rate
(@FAR 1E-6)
Performance Numbers
For Workstation or Server
Intel Core i7-12700K, 3.61GHz
* 1080p images, 1 face per image
* Face Match Time: Time taken to match two templates. For example, given a template extracted from a face image, to identify who is this person in a 1,000 people database takes 1,000 times matching
* FaceMe® - Windows 6.9 is used
Amazon EC2 G4g4dn.2xlarge
* 1080p images, 1 face per image. Each time batch process 16 images
* Face Match Time means Time taken to match two templates. For example, given a template extracted from a face image, to identify who is this person in a 1,000 people database takes 1,000 times matching
* Ubuntu 20 x64 is used
For Industrial PC
Execution Time (ms)
Detection + Extraction
Execution Time (ms)
Detection + Extraction
Execution Time (ms)
Detection + Extraction
* 1080p images, 1 face per image
* FaceMe® SDK - Windows 5.2 is used
* The "High Precision (DNN)" model is used for Face Detection
* For VPU case, test is done by running 960 images concurrently using Batch mode and measure the average time per image using FaceMe SDK 3.15
For Mobile Devices & IoT/ AIoT Devices
Qualcomm Snapdragon 660 (GPU)
Execution Time (ms)
Detection + Extraction
Execution Time (ms)Detection + Extraction
Qualcomm Snapdragon 660 (GPU)
Execution Time (ms)Detection + Extraction
Execution Time (ms)Detection + Extraction
Execution Time (ms)Detection + Extraction
* 720p images, 1 face per image
FaceMe® Technology
Features List
Detection, Extraction, Matching, Searching
Occlusion Detection, Emotion Detection
Landmark Detection, Medical Mask Detection
Face Attributes Recognition
Gender, Age, Emotion, Pose
H.264 (AVC), H.265 (HEVC)
Happy, Sad, Angry, Surprised, Neutral
3D Depth Anti-spoofing
IR+RGB Anti-spoofing
2D Camera Anti-spoofing (with or without random interactions)
Intel x64,
ARM64,
NVIDIA GPU, NVIDIA Jetson,
MediaTek SoC (i350, i500),
Qualcomm SoC (w/ GPU or DSP),
NXP i.MX8
FaceMe SDK Built-in,
Microsoft SQL, MySQL, Maria DB
Windows, Windows Server, Android, iOS
Ubuntu, Red Hat, CentOS, JetPack, Yocto*2
Intel® OpenVINO™,
NVIDIA® TensorRT / CUDA,
Qualcomm SNPE,
MediaTeK NeuroPilot,
Apple CoreML
* 1. Hardware decoder is required to support video formats
* 2. Yocto support is for specific OEM devices
Images/ Cameras Conditions
Face Size
(pixels between eyes)
5 images - Straight, Left/Right (15°~45°), Up (5°~30°), Down (15°~30°)
Anti-spoofing using 3D-Depth Camera
Himax SH430UH
iPhone X, and iPad Pro devices
Intel RealSense D415
Orbbec Astra Pro 3D
eYs3D EX8053
Union Ubio-X Face Premium series
Fangtec DCEA23-HK1-6R
TaisenTech H018-7168-V2
Supports almost all Android/iOS smart phones or tablets
* For 2D Camera column, The FaceMe Android/iOS is used with default settings.
* Tested with FaceMe 6.8
Fast Search Algorithm for an Enormous Database
Time to Search in Database (ms)
* Tested on an i7-7700K with 64GB memory PC, Windows 10 Pro.
* A database of 30 million faces would require 128 GB of memory.
Age, Gender, and Emotion Detection
5.8 y/o (Mean Average Error)
* Tested with 68,000 images.