ShiqiYu/libfacedetection: An open source library for face detection in images. The face detection speed can reach 1500FPS.

0

Get real time updates directly on you device, subscribe now.


This is an open source library for CNN-based face detection in images. The CNN model has been converted to stastic variales in C source files. The source code does not depend on any other libraries. What you need is just a C++ compiler. You can compile the source code under Windows, Linux, ARM and any platform with a C++ compiler.

SIMD instructions are used to speedup the detection. You can enable AVX2 if you use Intel CPU or NEON for ARM.

The model file has also been provided in directory ./models/.

examples/libfacedetectcnn-example.cpp shows how to use the library.

Examples

How to Compile

  • Please add -O3 to turn on optimizations when you compile the source code using g++.
  • Please choose ‘Maximize Speed/-O2’ when you compile the source code using Microsoft Visual Studio.

CNN-based Face Detection on Windows

Method Time FPS Time FPS
X64 X64 X64 X64
Single-thread Single-thread Multi-thread Multi-thread
OpenCV Haar+AdaBoost (640×480) 12.33ms 81.1
cnn (CPU, 640×480) 64.21ms 15.57 15.59ms 64.16
cnn (CPU, 320×240) 15.23ms 65.68 3.99ms 250.40
cnn (CPU, 160×120) 3.47ms 288.08 0.95ms 1052.20
cnn (CPU, 128×96) 2.35ms 425.95 0.64ms 1562.10
  • OpenCV Haar+AdaBoost runs with minimal face size 48×48
  • Face detection only, and no landmark detection included.
  • Minimal face size ~12×12
  • Intel(R) Core(TM) i7-7700 CPU @ 3.6GHz.

CNN-based Face Detection on ARM Linux (Raspberry Pi 3 B+)

Method Time FPS Time FPS
Single-thread Single-thread Multi-thread Multi-thread
cnn (CPU, 640×480) 512.04ms 1.95 174.89ms 5.72
cnn (CPU, 320×240) 123.47ms 8.10 42.13ms 23.74
cnn (CPU, 160×120) 27.42ms 36.47 9.75ms 102.58
cnn (CPU, 128×96) 17.78ms 56.24 6.12ms 163.50
  • Face detection only, and no landmark detection included.
  • Minimal face size ~12×12
  • Raspberry Pi 3 B+, Broadcom BCM2837B0, Cortex-A53 (ARMv8) 64-bit SoC @ 1.4GHz

Author

Contributors

  • Jia Wu
  • Shengyin Wu
  • Dong Xu

Acknowledgment

The work is partly supported by the Science Foundation of Shenzhen (Grant No. JCYJ20150324141711699).



Source

Get real time updates directly on you device, subscribe now.