FaceMe® | Optimization for NVIDIA | CyberLink
FaceMe®

FaceMe® - Optimized for NVIDIA® IoT Technologies

FaceMe®, a leader in facial recognition technologies, is designed for a wide range of IoT and AIoT applications and can be integrated with high-end servers, workstations and personal computers equipped with GPU to provide highly efficient performance. FaceMe® is optimized for TensorRT and supports a wide range of NVIDIA technologies, including CUDA and Jetson.

Customer Applications

Servers & Workstation

PCs

IoT Edge Devices

GPU & IoT Platform

Inference Engine & OS

  • NVIDIA® Metropolis Certified (T4)

    FaceMe® is certified by the NVIDIA Metropolis Partner Program, an edge-to-cloud platform connecting system integrators with software solutions. CyberLink's FaceMe® technology can uniquely utilize NVIDIA's GPU lineup to enhance overall performance while providing flexible settings to control speed, accuracy and CPU processing. Developers can build high-performance solutions for the most demanding throughput and latency requirements.

FaceMe® Performance

FaceMe® Security on Workstations with NVIDIA GPUs

Visitor Traffic
Hardware
NVIDIA RTX A6000

975

People per hour per channel

19 Channels

(Total 18,525 people/hour)
 

510

People per hour per channel

43 Channels

(Total 21,930 people/hour)
 

Visitor Traffic

975

People per hour per channel

Hardware
NVIDIA RTX A6000

19 Channels

(Total 18,525 people/hour)
 

Visitor Traffic

510

People per hour per channel

Hardware
NVIDIA RTX A6000

43 Channels

(Total 21,930 people/hour)
 

* 1080p images, Video Frame Rate: 15 FPS, Face Recognition Frame Rate: 10 FPS

* FaceMe® SECURITY 5.3 is used

* Tested Platforms: AMD EPYC 7502 CPU, NV RTX A6000, 32 GM Memory, Windows 10 Enterprise

Performance on VMS (Video Management Systems) with NVIDIA® Jetson

Jetson Nano
Jetson Xavier NX
Jetson Xavier
Extraction Model
VH
VH
UH
w / VH
UH
UH
w / VH
UH
Visitor Traffic
510 people per channel per hour
Number of Channels
4 ch.
16 ch.
11 ch.
4 ch.
13 ch.
7 ch.
Frames per second (fps) per channel
10.6 fps
10.4 fps
10.2 fps
10.6 fps
10.1 fps
11.9 fps
Jetson Nano
Extraction Model
VH
Visitor Traffic
510 people per channel per hour
Number of Channels
4 .ch
Frames per second (fps)
per chnannel
10.6 fps
Jetson Xavier NX
Extraction Model
VH
UH
w/ VH
UH
Visitor Traffic
510 people per channel per hour
Number of Channels
16 ch.
11 ch.
4 ch.
Frames per second (fps)
per channel
10.4 fps
10.2 fps
10.6 fps
Jetson Xavier
Extraction Model
UH
w/ VH
UH
Visitor Traffic
510 people per channel per hour
Number of Channels
13 ch.
7 ch.
Frames per second (fps)
per channel
10.1 fps
11.9 fps

* Tested with 1080p 15fps Video Streams

* Each channel has 510 people per hour walk through the camera. In the peak minutes of this hour, there’ll be 24 people pass by within 1 minute

* The "High Precision (DNN)" model is used for Face Detection

* FaceMe SDK-Jetson 5.3 is used on JetPack 4.4

Performance on NVIDIA® Jetson

Jetson Nano

Jetson Xavier NX

Jetson Xavier

FaceMe® Extraction Model
H
VH
VH
UH
UH
Execution time (ms)
Detection + extraction
35.8 ms
50.2 ms
13.7 ms
43.5 ms
26.5 ms
Frames per second (fps)
27.9 fps
19.9 fps
72.9 fps
23.0 fps
37.7 fps
Face Match Time (ms)
0.0075 ms
0.0075 ms
0.0017 ms
0.0037 ms
0.0024 ms

Jetson Nano

FaceMe® Extraction Model
H
VH
Execution time (ms)
Detection + extraction
35.8 ms
50.2 ms
Frames per second (fps)
27.9 fps
19.9 fps
Face Match Time (ms)
0.0075 ms
0.0075 ms

Jetson Xavier NX

FaceMe® Extraction Model
VH
UH
Execution Time (ms)
Detection + Extraction
13.7 ms
43.5 ms
Frames per second (fps)
72.9 fps
23.0 fps
Face Match Time (ms)
0.0017 ms
0.0037 ms

Jetson Xavier

FaceMe® Extraction Model
UH
Execution time (ms)
Detection + extraction
26.5 ms
Frames per second (fps)
37.7 fps
Face Match Time (ms)
0.0024 ms

* 1080p images, 1 face per image

* The "High Precision (DNN)" model is used for Face Detection

* FaceMe SDK-Jetson 5.3 is used on JetPack 4.4

For further information on CyberLink FaceMe® engine