继上文开箱后，本文主要依托爱芯元智官方的实例，进行官方YOLOV5模型的部署和测试。

一、环境搭建

由于8核A55的SoC，加上目前Debian OS的工具齐全，所以决定直接在板上编译程序。

root@maixbox:~# lscpu
Architecture:            aarch64
  CPU op-mode(s):        32-bit, 64-bit
  Byte Order:            Little Endian
CPU(s):                  8
  On-line CPU(s) list:   0-7
Vendor ID:               ARM
  Model name:            Cortex-A55
    Model:               0
    Thread(s) per core:  1
    Core(s) per cluster: 8
    Socket(s):           -
    Cluster(s):          1
    Stepping:            r2p0
    CPU(s) scaling MHz:  100%
    CPU max MHz:         1700.0000
    CPU min MHz:         1200.0000
    BogoMIPS:            48.00
    Flags:               fp asimd evtstrm crc32 atomics fphp asimdhp cpuid asimdrdm lrcpc dcpop asimddp

开发工具什么的全部装上，apt install build-essential libopencv-dev cmake。看看gcc版本。

root@maixbox:~# gcc -v
Using built-in specs.
COLLECT_GCC=gcc
COLLECT_LTO_WRAPPER=/usr/lib/gcc/aarch64-linux-gnu/12/lto-wrapper
Target: aarch64-linux-gnu
Configured with: ../src/configure -v --with-pkgversion='Debian 12.2.0-14' --with-bugurl=file:///usr/share/doc/gcc-12/README.Bugs --enable-languages=c,ada,c++,go,d,fortran,objc,obj-c++,m2 --prefix=/usr --with-gcc-major-version-only --program-suffix=-12 --program-prefix=aarch64-linux-gnu- --enable-shared --enable-linker-build-id --libexecdir=/usr/lib --without-included-gettext --enable-threads=posix --libdir=/usr/lib --enable-nls --enable-clocale=gnu --enable-libstdcxx-debug --enable-libstdcxx-time=yes --with-default-libstdcxx-abi=new --enable-gnu-unique-object --disable-libquadmath --disable-libquadmath-support --enable-plugin --enable-default-pie --with-system-zlib --enable-libphobos-checking=release --with-target-system-zlib=auto --enable-objc-gc=auto --enable-multiarch --enable-fix-cortex-a53-843419 --disable-werror --enable-checking=release --build=aarch64-linux-gnu --host=aarch64-linux-gnu --target=aarch64-linux-gnu
Thread model: posix
Supported LTO compression algorithms: zlib zstd
gcc version 12.2.0 (Debian 12.2.0-14)

二、源码下载&编译

首先，git clone https://github.com/AXERA-TECH/ax-samples.git下载源码到本地。
然后，指定芯片为AX650，cmake生成makefile。

cd ax-samples 
mkdir build && cd build
cmake -DBSP_MSP_DIR=/soc/ -DAXERA_TARGET_CHIP=ax650 ..

第三步，make -j8，既然8核，那就-j8全速。

过一会等编译完成。

最后，make install。可以看到生成的可执行示例存放在build/install/ax650/ 路径下。

其中很多案例程序，因为智能教室需要清点人数，所以选择了YOLOV5和YOLOV7_TINY face两个demo。

三、下载模型

爱芯元智官方自己搞了个ModelZoo，类似于AMD Vitis AI 的Vitis AI Model Zoo工具，主要是提供AXERA芯片平台的通用AI模型（具体baidu盘链接为 https://pan.baidu.com/s/1CCu-oKw8jUEg2s3PEhTa4g?pwd=xq9f ），直接下载下来使用。

因为本文需要部署人脸识别模型，所以下载了yolov7-tiny-face.axmodel和yolov5s-face.axmodel两个模型文件。

四、测试

选择了一张不错的关于教室课堂的测试图片。

（一）YOLOV5人脸识别

root@maixbox:~/ax-samples/build/install/ax650# ./ax_yolov5_face -m /root/yolov5s-face.axmodel -i /root/CLASS.jpg
--------------------------------------
model file : /root/yolov5s-face.axmodel
image file : /root/CLASS.jpg
img_h, img_w : 640 640
--------------------------------------
Engine creating handle is done.
Engine creating context is done.
Engine get io info is done.
Engine alloc io is done.
Engine push input is done.
--------------------------------------
post process cost time:0.39 ms
--------------------------------------
Repeat 1 times, avg time 7.80 ms, max_time 7.80 ms, min_time 7.80 ms
--------------------------------------
detection num: 5
 0:  87%, [ 145,  113,  226,  218], face
 0:  87%, [ 338,  160,  419,  254], face
 0:  81%, [ 519,  138,  561,  184], face
 0:  81%, [ 604,  250,  647,  294], face
 0:  77%, [ 495,  265,  536,  311], face
--------------------------------------

五张人脸识别耗时7.8ms，识别准确率较高，识别后的图片如下：

（二）YOLOV7_TINY人脸识别

root@maixbox:~/ax-samples/build/install/ax650# ./ax_yolov7_tiny_face -m /root/yolov7-tiny-face.axmodel -i /root/CLASS.jpg
--------------------------------------
model file : /root/yolov7-tiny-face.axmodel
image file : /root/CLASS.jpg
img_h, img_w : 640 640
--------------------------------------
Engine creating handle is done.
Engine creating context is done.
Engine get io info is done.
Engine alloc io is done.
Engine push input is done.
--------------------------------------
post process cost time:0.84 ms
--------------------------------------
Repeat 1 times, avg time 8.98 ms, max_time 8.98 ms, min_time 8.98 ms
--------------------------------------
detection num: 5
 0:  88%, [ 147,  115,  224,  218], face
 0:  88%, [ 344,  159,  415,  252], face
 0:  84%, [ 520,  137,  560,  185], face
 0:  84%, [ 605,  250,  644,  296], face
 0:  74%, [ 498,  267,  535,  311], face
--------------------------------------

五张人脸识别耗时8.98ms，识别准确率总体比YOLOV5高，识别后的图片如下：

五、总结

得益于官方案例，开发者可以很方便的在爱芯元智SoC硬件平台上部署常见的深度学习算法模型，方便开发者快速评估和适配业务。