【EASY EAI Nano人工智能开发套件试用体验】EASY EAI Nano人工智能开发套件连接屏幕和摄像头及跑分评测

评测EASY EAI Nano人工智能开发套件连接屏幕和摄像头，并做个 编译ncnn跑分，首先将开发板套件的几大硬件做连接，整个连接过程还是很方便的，吐槽一下，主板和5寸屏幕的连接线有点不太方便，组装好的化，连接线刚好挡住TF卡槽，取和装TF卡都不方便。
再主板和5寸屏幕的连接线拆卸几次后有点接触不良，屏幕会出现无法点亮的情况，重现安装连线可以解决，但取和装TF卡有一点点不方便。

下面编译ncnn，和做个跑分测试
编译 ncnn 的跑分工具看看模型推理耗时

1. 先从百度网盘下载工具链 03.编译器/rv1126_rv1109_compiler_xxxxxxxx.tar.gz

2. 工具链解压到本地，比如 /home/nihui/osd/opt/rv1126_rv1109_sdk

3. git clone https://github.com/Tencent/ncnn.git 下载最新版本 ncnn

4. 新建 ncnn/toolchains/rv1126.toolchain.cmake

set(CMAKE_SYSTEM_NAME Linux)
set(CMAKE_SYSTEM_PROCESSOR arm)

set(RV1126_ROOT_PATH "/home/nihui/osd/opt/rv1126_rv1109_sdk/prebuilts/gcc/linux-x86/arm/gcc-arm-8.3-2019.03-x86_64-arm-linux-gnueabihf")

set(CMAKE_C_COMPILER "${RV1126_ROOT_PATH}/bin/arm-linux-gnueabihf-gcc")
set(CMAKE_CXX_COMPILER "${RV1126_ROOT_PATH}/bin/arm-linux-gnueabihf-g++")

set(CMAKE_FIND_ROOT_PATH "${RV1126_ROOT_PATH}/arm-linux-gnueabihf")

set(CMAKE_FIND_ROOT_PATH_MODE_PROGRAM NEVER)
set(CMAKE_FIND_ROOT_PATH_MODE_LIBRARY ONLY)
set(CMAKE_FIND_ROOT_PATH_MODE_INCLUDE ONLY)

set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -march=armv7-a -mfloat-abi=hard -mfpu=neon")
set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -march=armv7-a -mfloat-abi=hard -mfpu=neon")

# cache flags
set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS}" CACHE STRING "c flags")
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS}" CACHE STRING "c++ flags")

5. 交叉编译rv1126的ncnn库和跑分工具

mkdir build
cd build
cmake -DCMAKE_TOOLCHAIN_FILE=../toolchains/rv1126.toolchain.cmake ..
make -j4

6. 上传benchncnn和param文件到小板子上

scp ncnn/build/benchmark/benchncnn root@target_ip:/root/
scp ncnn/benchmark/*.param root@target_ip:/root/

7. 因为系统里缺少了 libgomp.so.1 库，无法直接运行 benchncnn，于是也上传到小板子上

scp /home/nihui/osd/opt/rv1126_rv1109_sdk/prebuilts/gcc/linux-x86/arm/gcc-arm-8.3-2019.03-x86_64-arm-linux-gnueabihf/arm-linux-gnueabihf/lib/libgomp.so.1 root@target_ip:/root/
scp /home/nihui/osd/opt/rv1126_rv1109_sdk/prebuilts/gcc/linux-x86/arm/gcc-arm-8.3-2019.03-x86_64-arm-linux-gnueabihf/arm-linux-gnueabihf/lib/libgomp.so.1.0.0 root@target_ip:/root/

8. 运行benchncnn，结果符合 Cortex-A7 1.5GHz 4核心水平，手指碰碰芯片一点也不烫

[root@EASY-EAI-NANO:~]# LD_LIBRARY_PATH=. ./benchncnn 4 1 0 -1 0
loop_count = 4
num_threads = 1
powersave = 0
gpu_device = -1
cooling_down = 0
          squeezenet  min =  347.29  max =  349.57  avg =  348.68
     squeezenet_int8  min =  243.58  max =  246.40  avg =  244.66
           mobilenet  min =  587.28  max =  590.97  avg =  589.45
      mobilenet_int8  min =  355.43  max =  356.95  avg =  356.13
        mobilenet_v2  min =  384.40  max =  388.06  avg =  386.81
        mobilenet_v3  min =  313.53  max =  316.49  avg =  315.00
          shufflenet  min =  195.54  max =  196.91  avg =  196.09
       shufflenet_v2  min =  189.91  max =  191.17  avg =  190.60
             mnasnet  min =  378.74  max =  381.65  avg =  380.00
     proxylessnasnet  min =  438.49  max =  441.41  avg =  440.19
     efficientnet_b0  min =  642.24  max =  643.60  avg =  642.99
   efficientnetv2_b0  min =  770.34  max =  776.84  avg =  773.29
        regnety_400m  min =  487.86  max =  488.81  avg =  488.23
           blazeface  min =   58.51  max =   59.56  avg =   59.01
           googlenet  min = 1125.79  max = 1129.31  avg = 1127.91
      googlenet_int8  min =  766.64  max =  775.92  avg =  772.11
            resnet18  min = 1026.52  max = 1034.20  avg = 1030.92
       resnet18_int8  min =  599.47  max =  601.88  avg =  600.49
             alexnet  min =  739.84  max =  740.37  avg =  740.08
            resnet50  min = 2932.26  max = 2940.41  avg = 2935.45
       resnet50_int8  min = 1704.34  max = 1709.17  avg = 1706.85
      squeezenet_ssd  min =  808.41  max =  810.73  avg =  809.45
 squeezenet_ssd_int8  min =  588.13  max =  592.90  avg =  589.99
       mobilenet_ssd  min = 1236.99  max = 1240.53  avg = 1239.07
  mobilenet_ssd_int8  min =  717.88  max =  719.66  avg =  718.53
      mobilenet_yolo  min = 2859.88  max = 2870.28  avg = 2863.96
  mobilenetv2_yolov3  min = 1422.87  max = 1432.51  avg = 1428.08
         yolov4-tiny  min = 1815.68  max = 1819.83  avg = 1818.05
           nanodet_m  min =  451.71  max =  453.34  avg =  452.66
    yolo-fastest-1.1  min =  206.27  max =  208.01  avg =  206.81
      yolo-fastestv2  min =  170.55  max =  172.22  avg =  171.54

[root@EASY-EAI-NANO:~]# LD_LIBRARY_PATH=. ./benchncnn 4 4 0 -1 0
loop_count = 4
num_threads = 4
powersave = 0
gpu_device = -1
cooling_down = 0
          squeezenet  min =  105.16  max =  107.44  avg =  106.40
     squeezenet_int8  min =   77.16  max =   78.18  avg =   77.57
           mobilenet  min =  163.82  max =  164.85  avg =  164.44
      mobilenet_int8  min =   97.31  max =   97.56  avg =   97.41
        mobilenet_v2  min =  120.65  max =  123.63  avg =  121.44
        mobilenet_v3  min =   96.38  max =   96.98  avg =   96.74
          shufflenet  min =   68.79  max =   69.17  avg =   69.03
       shufflenet_v2  min =   69.65  max =   70.27  avg =   69.96
             mnasnet  min =  114.84  max =  115.03  avg =  114.91
     proxylessnasnet  min =  128.13  max =  129.04  avg =  128.44
     efficientnet_b0  min =  185.84  max =  186.70  avg =  186.38
   efficientnetv2_b0  min =  221.02  max =  221.52  avg =  221.22
        regnety_400m  min =  176.50  max =  176.90  avg =  176.76
           blazeface  min =   21.35  max =   22.02  avg =   21.72
           googlenet  min =  320.95  max =  322.98  avg =  321.78
      googlenet_int8  min =  223.19  max =  223.78  avg =  223.48
            resnet18  min =  320.52  max =  323.48  avg =  322.19
       resnet18_int8  min =  168.64  max =  169.50  avg =  169.04
             alexnet  min =  196.96  max =  197.35  avg =  197.11
            resnet50  min =  822.21  max =  830.47  avg =  824.50
       resnet50_int8  min =  473.53  max =  474.48  avg =  474.06
      squeezenet_ssd  min =  268.59  max =  269.26  avg =  268.83
 squeezenet_ssd_int8  min =  196.17  max =  196.82  avg =  196.42
       mobilenet_ssd  min =  336.65  max =  337.49  avg =  337.12
  mobilenet_ssd_int8  min =  198.52  max =  199.05  avg =  198.71
      mobilenet_yolo  min =  765.30  max =  768.77  avg =  767.09
  mobilenetv2_yolov3  min =  419.66  max =  420.69  avg =  420.06
         yolov4-tiny  min =  560.99  max =  568.58  avg =  564.52
           nanodet_m  min =  169.19  max =  170.39  avg =  169.96
    yolo-fastest-1.1  min =   81.48  max =   82.68  avg =   81.94
      yolo-fastestv2  min =   66.36  max =   67.32  avg =   66.87

测试结果还是很不错的。