Coder Social home page Coder Social logo

kingrain213 / zqcnn-mtcnn-vs-libfacedetection Goto Github PK

View Code? Open in Web Editor NEW

This project forked from zuoqing1988/zqcnn-mtcnn-vs-libfacedetection

0.0 2.0 0.0 92.55 MB

对比ZQCNN-MTCNN与libfacedetection

CMake 0.27% C++ 88.30% Shell 0.03% C 11.35% Makefile 0.01% Java 0.03% Objective-C 0.01% Objective-C++ 0.01%

zqcnn-mtcnn-vs-libfacedetection's Introduction

ZQCNN-MTCNN-vs-libfacedetection

说明1:libfacedetection的caffe模型转成ZQCNN格式之后能速度接近翻倍,请前往ZQCNN项目,在SampleSSD里可以试用

说明2:ZQCNN在不断更新,此版本会有一些小BUG,在不断修复,请下载最新版本

ZQCNN:下载时间2019-03-25 09:00

依赖库:

windows:mkl, opencv3.4.2

arm-linux: openblas, opencv3.4.2

libfacedetection:下载时间2019-03-25 09:00

(一) WINDOWS下对比 (测试机器为E5-1650V4 3.6GHz)

ZQCNN用.sln打开

libfacedetection用cmake-gui配置,勾选avx2和DEMO,不勾选int8和neon,配置之后用vs打开,注意在facedetection、facedetection_shared、fdt_demo三个项目的项目属性->C++->代码生成里开启AVX2和快速浮点。

(1)对比keliamoniz1.jpg

输入尺寸 ZQCNN单线程 libfacedetection单线程 备注
640x480 27ms 75ms nIter=1000
320x240 9.0ms 18ms nIter=1000
160x120 2.8ms 4.0ms nIter=1000
128x96 2.0ms 2.6ms nIter=10000

(2)对比4.jpg

输入尺寸 ZQCNN单线程 ZQCNN找到个数 libfacedetection单线程 libfacedetection找到个数 备注
640x480 30ms 4 75ms 4 nIter=1000
320x240 10ms 4 18ms 4 nIter=1000
160x120 4.7ms 4 4.0ms 4 nIter=1000
128x96 3.5ms 4 2.6ms 3 nIter=10000

(3)对比其他图片

图片名 输入尺寸 ZQCNN单线程 ZQCNN找到个数 libfacedetection单线程 libfacedetection找到个数 备注
1.jpg 1920x1080 165ms 6 522ms 10 (2虚检) nIter=100
2.jpg 3164x1440 354ms 44 (2虚检) 1200ms 42 (很多框不准) nIter=10
test.jpg 1920x1080 200ms 60 526ms 50 nIter=100
test2.jpg 2048x1217 240ms 82 656ms 50 nIter=50
test11.jpg 5520x3680 1800ms 844 5230ms 50 nIter=10
test12.jpg 1280x720 85ms 14 238ms 14 nIter=100

(二) ARM-LINUX下对比 (测试机器为firefly rk3399)

ZQCNN: 先编译OpenBLAS, OpenCV3.4.2, 然后编译ZQCNN, 使用命令cmake .. -DSIMD_ARCH_TYPE=arm64 -DBLAS_TYPE=openblas_zq_gemm 具体参见ZQCNN项目的README

libfacedetection: 命令cmake .. -DENABLE_NEON=ON -DCMAKE_BUILD_TYPE=RELEASE

(1)对比keliamoniz1.jpg

输入尺寸 ZQCNN单A72 ZQCNN单A53 libfacedetection单A72 libfacedetection单A53 备注
640x480 129ms 327ms 341ms 628ms nIter=100
320x240 37.7ms 92.8ms 84ms 152ms nIter=100
160x120 11.5ms 29.2ms 19.5ms 35.8ms nIter=500
128x96 8.77ms 22.2ms 13.0ms 23.9ms nIter=1000

(2)对比4.jpg

输入尺寸 ZQCNN单A72 ZQCNN单A53 ZQCNN找到个数 libfacedetection单A72 libfacedetection单A53 libfacedetection找到个数 备注
640x480 133ms 340ms 4 315ms 598ms 4 nIter=100
320x240 42.5ms 108ms 4 77.5ms 144ms 4 nIter=100
160x120 19.7ms 50.9ms 4 18.0ms 33.7ms 4 nIter=500
128x96 13.5ms 35.8ms 4 12.0ms 22.6ms 3 nIter=1000

(3)对比其他图片

图片名 输入尺寸 ZQCNN单A72 ZQCNN单A53 ZQCNN找到个数 libfacedetection单A72 libfacedetection单A53 libfacedetection找到个数 备注
1.jpg 1920x1080 845ms 2230ms 6 2300ms 4265ms 10 (2虚检) nIter=10
test.jpg 1920x1080 940ms 2490ms 60 2307ms 4266ms 50 nIter=10
test2.jpg 2048x1217 1210ms 3150ms 82 2800ms 5185ms 50 nIter=10
test12.jpg 1280x720 412ms 1110ms 14 1034ms 1896ms 14 nIter=10

zqcnn-mtcnn-vs-libfacedetection's People

Contributors

zuoqing1988 avatar

Watchers

 avatar  avatar

Recommend Projects

  • React photo React

    A declarative, efficient, and flexible JavaScript library for building user interfaces.

  • Vue.js photo Vue.js

    🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.

  • Typescript photo Typescript

    TypeScript is a superset of JavaScript that compiles to clean JavaScript output.

  • TensorFlow photo TensorFlow

    An Open Source Machine Learning Framework for Everyone

  • Django photo Django

    The Web framework for perfectionists with deadlines.

  • D3 photo D3

    Bring data to life with SVG, Canvas and HTML. 📊📈🎉

Recommend Topics

  • javascript

    JavaScript (JS) is a lightweight interpreted programming language with first-class functions.

  • web

    Some thing interesting about web. New door for the world.

  • server

    A server is a program made to process requests and deliver data to clients.

  • Machine learning

    Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.

  • Game

    Some thing interesting about game, make everyone happy.

Recommend Org

  • Facebook photo Facebook

    We are working to build community through open source technology. NB: members must have two-factor auth.

  • Microsoft photo Microsoft

    Open source projects and samples from Microsoft.

  • Google photo Google

    Google ❤️ Open Source for everyone.

  • D3 photo D3

    Data-Driven Documents codes.