EEPW论坛

飞凌RK3568开发板试用体验】RKNN模型推理测试：https://bbs.elecfans.com/jishu_2321748_1_1.html【飞凌RK3568开发板试用体验】使用RKNN Toolkit2转换Pytorch模型：https://bbs.elecfans.com/jishu_2328460_1_1.html

今天我们主要使用厂商自己提供的模型来测试其RKNN推理能力。

首先，需要安装厂商提供的LuckFox
Pico SDK：https://github.com/LuckfoxTECH/luckfox-pico/。然后，安装瑞芯微的rknpu2：https://github.com/rockchip-linux/rknpu2。如果想转换自己的模型，还需要安装瑞芯微的RKNN Toolkit2：https://github.com/rockchip-linux/rknn-toolkit2。

在rknpu2的example目录下可以找到RV1106_RV1103目录，其中包括2个可以直接用于LuckFox Pico开发板的例程。

编译例程前需要设置如下环境变量：

export RK_RV1106_TOOLCHAIN=~/luckfox-pico/tools/linux/toolchain/arm-rockchip830-linux-uclibcgnueabihf/bin/arm-rockchip830-linux-uclibcgnueabihf

然后执行examples/RV1106_RV1103/rknn_yolov5_demo目录下的build-linux_RV1106.sh脚本。该脚本将编译例程：

-- The C
compiler identification is GNU 8.3.0

-- The CXX
compiler identification is GNU 8.3.0

-- Check for
working C compiler:
/home/test/luckfox-pico/tools/linux/toolchain/arm-rockchip830-linux-uclibcgnueabihf/bin/arm-rockchip830-linux-uclibcgnueabihf-gcc

-- Check for
working C compiler:
/home/test/luckfox-pico/tools/linux/toolchain/arm-rockchip830-linux-uclibcgnueabihf/bin/arm-rockchip830-linux-uclibcgnueabihf-gcc
-- works

-- Detecting C
compiler ABI info

-- Detecting C
compiler ABI info - done

-- Detecting C
compile features

-- Detecting C
compile features - done

-- Check for
working CXX compiler:
/home/test/luckfox-pico/tools/linux/toolchain/arm-rockchip830-linux-uclibcgnueabihf/bin/arm-rockchip830-linux-uclibcgnueabihf-g++

-- Check for
working CXX compiler:
/home/test/luckfox-pico/tools/linux/toolchain/arm-rockchip830-linux-uclibcgnueabihf/bin/arm-rockchip830-linux-uclibcgnueabihf-g++
-- works

-- Detecting CXX
compiler ABI info

-- Detecting CXX
compiler ABI info - done

-- Detecting CXX
compile features

-- Detecting CXX
compile features - done

-- Configuring
done

-- Generating
done

-- Build files
have been written to:
/mnt/ok/rknpu2-master/examples/RV1106_RV1103/rknn_yolov5_demo/build/build_linux_arm

Scanning
dependencies of target rknn_yolov5_demo

[ 66%] Building
CXX object CMakeFiles/rknn_yolov5_demo.dir/src/postprocess.cc.o

[ 66%] Building
CXX object CMakeFiles/rknn_yolov5_demo.dir/src/main.cc.o

[100%] Linking
CXX executable rknn_yolov5_demo

[100%] Built
target rknn_yolov5_demo

[100%] Built
target rknn_yolov5_demo

Install the
project...

-- Install
configuration: ""

-- Installing:
/mnt/ok/rknpu2-master/examples/RV1106_RV1103/rknn_yolov5_demo/install/rknn_yolov5_demo_Linux/./rknn_yolov5_demo

-- Installing:
/mnt/ok/rknpu2-master/examples/RV1106_RV1103/rknn_yolov5_demo/install/rknn_yolov5_demo_Linux/lib/librknnmrt.so

-- Installing:
/mnt/ok/rknpu2-master/examples/RV1106_RV1103/rknn_yolov5_demo/install/rknn_yolov5_demo_Linux/.//model

-- Installing:
/mnt/ok/rknpu2-master/examples/RV1106_RV1103/rknn_yolov5_demo/install/rknn_yolov5_demo_Linux/.//model/bus.jpg

-- Installing:
/mnt/ok/rknpu2-master/examples/RV1106_RV1103/rknn_yolov5_demo/install/rknn_yolov5_demo_Linux/.//model/coco_80_labels_list.txt

-- Installing:
/mnt/ok/rknpu2-master/examples/RV1106_RV1103/rknn_yolov5_demo/install/rknn_yolov5_demo_Linux/.//model/RV1106

-- Installing:
/mnt/ok/rknpu2-master/examples/RV1106_RV1103/rknn_yolov5_demo/install/rknn_yolov5_demo_Linux/.//model/RV1106/yolov5s-640-640.rknn

连接板子的usb口到电脑，使用ADB命令将install目录下的整个demo目录复制到到 /userdata:

adb push install/rknn_yolov5_demo_Linux /userdata/

然后在板子上进入ADB shell运行程序。

adb shell

cd /userdata/rknn_yolov5_demo_Linux/

chmod 777 rknn_yolov5_demo

export LD_LIBRARY_PATH=/userdata/rknn_yolov5_demo_Linux/lib

./rknn_yolov5_demo model/RV1106/yolov5s-640-640.rknn model/bus.jpg

程序运行结果如下图：

我们用OpenCV绘制了识别结果，如下图。

识别的过程耗时100ms，速度还是相当快的。