FaceMe® | Optimization for Intel | CyberLink
FaceMe®

FaceMe®

Optimization for Intel® IoT Platforms

CyberLink is a partner of Intel® IoT Solutions Alliance. FaceMe® AI facial recognition is optimized for Intel® IoT technologies & platforms, including OpenVINO, Movidius and RealSense, enhancing powerful, flexible and highly secured facial recognition engines, which can be deployed on a broad range of IoT applications and scenarios.

Optimization for Intel® IoT Technologies

Our partnership with Intel removes entry barriers, facilitating the adoption of a wide array of edge-based AI facial recognition.

5.4x
Faster with OpenVINO

Optimization that provides performance enhancements across the board.

17.4x
Faster with Movidius VPU

A flexible solution that enables access to several low-cost IoT and AIoT use cases.

100%
Spoofing-attack Prevention

High-performing 3D anti-spoofing capabilities, crucial for a number of authentication scenarios.

5.4x Acceleration with Intel® OpenVINO

  • FaceMe® is highly optimized for OpenVINO. With OpenVINO acceleration enabled, FaceMe®’s facial recognition performance can increase by 5.4 times.

    • Benchmark Platform
    • CPU: Intel Core i7-7700K
    • OS: Windows 10 x64
    • Frame Size (Pixel): 1,920 x 1,080
    Download Whitepaper
    Learn how FaceMe® delivers high-performance vision with OpenVINO

Optimization for Intel® Movidius VPU

Intel® Movidius VPU (Vision Processing Unit) enables AI models on small factor, low power devices. FaceMe® runs 17.4x faster with Movidius VPU when compared to a Celeron® CPU, in addition to saving 72% of CPU usage.

FaceMe® Optimization on Single Movidius VPU

Test Platform
Advantech DS-580
Advantech
DS-580
  • Advantech DS-580
  • Intel® Celeron® J3355, 8 GB Memory, Logitech 920
  • Face Detection model used: High Precision Model (DNN)
  • Face Extraction model used: Ultra High Precision Model (UH)
FaceMe® Ultra Precision Model
Intel® CPU
(Celeron® J3355)
Intel® Movidius VPU
(Myriad X)
Process Time
4,078 ms
234 ms
17.4x
Speed
Average CPU Usage
90-97%
25-28%
72%
Less CPU Usage
FaceMe® Ultra Precision Model
FaceMe® CPU (Celeron® J3355)
Process Time
Average CPU Usage
4,078 ms
90-97%
Intel® Movidius VPU
(Myriad X)
Process Time
Average CPU Usage
234 ms
25-28%
17.4x
Speed
72%
Less CPU Usage

FaceMe® Optimization on Multiple Movidius VPU

Test Platform
iEi Mustang-V100-MX4/MX8
iEi
Mustang-V100-MX4/MX8
  • Intel® Celeron® G4920, 4 GB Memory
  • Intel NCS2, IEI Mustang-V100-MX4 & Mustang-V100-MX8
  • Face Detection model: High Precision Model (DNN)
  • Face Extraction model: Ultra High Precision Model (UH3)
Number of VPUs
Process Time
Performance Gain
1x Movidius VPU
Intel® Celeron® G4920 + Intel NCS2
415 ms
4x Movidius VPU
Intel® Celeron® G4920 + iEi Mustang-V100-MX4
115 ms
3.6x Speed
compared to 1x VPU
8x Movidius VPU
Intel® Celeron® G4920 + iEi Mustang-V100-MX8
69.2 ms
6x Speed
compared to 1x VPU
Number of VPUs
1x Movidius VPU
Intel® Celeron® G4920 + Intel NCS2
4x Movidius VPU
Intel® Celeron® G4920 + iEi Mustang-V100-MX4
8x Movidius VPU
Intel® Celeron® G4920 + iEi Mustang-V100-MX8
Process Time
415 ms
115 ms
69.2 ms
Performance Gain
3.6x
speed compared to 1x VPU
6x
speed compared to 1x VPU

Secure Anti-Spoofing with Intel® RealSense

RealSense™ D415

RealSense D415

Through its integration with Intel® RealSense camera, FaceMe® can extract a precise depth map, enabling highly secure protection against biometric fraud such as replay and print attacks. Benchmarks established a true positive rate (detecting a real person) of 98.2% and a true negative rate (detecting spoofing attempt) of 100%.

Contact
Our Sales Team

Contact the FaceMe Team