Sign In
4
Introducing Award-winning Creative Editing Solutions
FaceMe® The World’s Top Cross-Platform AI Facial Recognition Engine

FaceMe® Accuracy

Ultra-high Precision Model
High Precision Model
True Acceptance Rate (@FAR 1E-4)
99.50%
98.25%
True Acceptance Rate (@FAR 1E-6)
98.50%
94.12%
Model Size (MB)
250
4

* Tested with more than 1 million faces.

* In the most recent MegaFace Challenge, CyberLink FaceMe® achieved an accuracy rate at 98.41% and is ranked #12 in the world. (September, 2018)

FaceMe® Accuracy Report from NIST 2019 (FNMR to FMR)

Note: For visa images, detection error tradeoff (DET) characteristics showing false non-match rate vs. false match rate plotted parametrically on threshold, T. The scales are logarithmic in order to show many decades of FMR.

Speed for High Precision Model

Linux Server
Windows PC
Android
iOS
Intel XEON E5-2630V4 10 cores
Intel i7-7700K 4 cores
Snapdragon 835(On Pixel 2)
A12X (On iPad Pro 11” 2018)
Detection + Extraction Time (ms)
13 ms
12 ms
35 ms
14 ms
Frames per second
77 FPS
83 FPS
28 FPS
70 FPS
Face Match Time (ms)
0.0005 ms
0.0005 ms
0.0005 ms
0.0007 ms
Linux Server
Windows PC
Intel XEON E5-2630V4 10 cores
Intel i7-7700K 4 cores
Detection + Extraction Time (ms)
13 ms
12 ms
Frames per second
77 FPS
83 FPS
Face Match Time (ms)
0.0005 ms
0.0005 ms
Android
iOS
Snapdragon 835 (On Pixel 2)
A12X (On iPad Pro 11” 2018)
Detection + Extraction Time (ms)
35 ms
14 ms
Frames per second
28 FPS
70 FPS
Face Match Time (ms)
0.0005 ms
0.0007 ms

* Face Match Time: Time taken to match two templates.

FaceMe® Technology

Features List
Face Detection
Yes
Face Landmark
Yes
Face Recognition
Yes
Face Tracking
Yes
Face Attributes Recognition
Gender, Age, Emotion, Pose
Real-time Video Support
Yes
Emotion Detection
Happy, Sad, Angry, Surprised, Neutral
Images Pre-Processing
Lighting Enhancement, Edge Sharpening, Upsampling
Anti-Spoofing with 3D-Cameras
Yes
Anti-Spoofing with 2D-Cameras (with user interactions)
Yes
Anti-Spoofing with 2D-Cameras (without user interactions)
Yes
AI Inference Engines
TensorFlow, OpenVINO, CoreML

Images/Cameras Conditions

Minimum
Recommended
Face Pose – Yaw
< 60°
< 45°
Face Pose – Pitch
< 50°
< 30°
Face Pose – Roll
< 60°
< 45°
Lighting
450 lux
550 lux
Face Size (pixels between eyes)
36 pixels
50 pixels
Face Enrollment
1 image
5 images (Straight, 15° < Left/Right < 45°, 5° < Up < 30°, 15° < Down < 30°)

Recall Rate for Anti-spoofing Using 3D-Depth Camera

True Positive Rate
98.2%
True Negative Rate
100%

* Using Orbbec 3D Depth Camera (Active Stereo), over 17,000 test cases including real person, printed photo, video attacks, etc..

Fast Search Algorithm for Very Large Face Database

People in Database
Time to Search in Database (ms)
Recall Rate
One-by-one Compare
Fast Search
One-by-one Compare
Fast Search
1 Million People
1,860 ms
1.2 ms
100%
99.94%
6 Million People
11,200 ms
1.6 ms
100%
99.61%
Time to Search in Database (ms)
One-by-one Compare
Fast Search
1 Million People
1,860 ms
1.2 ms
6 Million People
11,200 ms
1.6 ms
Recall Rate
One-by-one Compare
Fast Search
1 Million People
100%
99.94%
6 Million People
100%
99.61%

* With Fast Search algorithm, FaceMe can search a face within a 6 million person database for about 1.6 ms with a recall rate up to 99.61%. The recall and performance tradeoff can be adjusted to meet different requirements.

* Tested on a i7-7700K with 24GB memory PC. A database of 30 million faces would require 128GB memory.

Accuracy of Face Attributes Detection

Detection Attributes
Accuracy Rate
Gender
98%
Emotion
Up to 86%
Age
5.8 y/o (Mean Average Error)

* Tested with 68,000 images.

Have questions or need to know more about CyberLink technology solutions?

Contact Us