rk3399pro上安装rknnplugin的各模型测试
# DIY operators rknnplugin
环境：RK3399pro + fedora
[](./LICENSE)
1. 下载Tengine [Tengine in GitHub]([email protected]:8022/internal/tengine.git)
git clone [email protected]:8022/internal/tengine.git
cd Tengine
2. Checkout tag v1.3.2
git checkout -b v1.3.2 v1.3.2
3. Build Protobuf
git clone [email protected]:protocolbuffers/protobuf.git
cd protobuf
git checkout -b v3.5.0 v3.5.0
./autogen.sh
./configure --with-protoc=protoc
make -j4
make install
4. Build opencv
git clone [email protected]:opencv/opencv.git
#(https://github.com/opencv/opencv.git)
git checkout -b 3.4.0 3.4.0
cd opencvcp build_opencv.sh .
mkdir build
cd build
../build_opencv.sh
make -j4
sudo make install5. Build Tengine
mkdir build
./linux_build.sh default_config/arm64_linux_native.config
make -j4
make install6. Download rknn SDK. see [Toybrick]
此处我下载的是rknn-api-0.9.9，放置在home目录下备用。
7. Build rknnplugin
cd #此处是你的rknnplugin根目录
mkdir build
cd build
vim ../linux_build.sh 需要将tengin的根路径、将要安装至哪里的路径、rknn SDK的根路径 指定对。
../linux_build.sh
cmake ../
make -j4
make install若有一些找不到路径的错误，可以尝试以下修改：
1.在rknnplugin-master/CMakeLists修改
set(LIB_PATH ${RKNN_SDK_PATH}/rknn_api/arm/lib64)
include_directories(${RKNN_SDK_PATH}/rknn_api/arm/include)
为
set(LIB_PATH ${RKNN_SDK_PATH}/rknn_api/arm/lib64)
include_directories(${RKNN_SDK_PATH}/rknn_api/arm/include)
2.在rknnplugin-master/rknn/CMakeLists.txt 中修改
include_directories(${RKNN_SDK_PATH}/rknn_api/arm/include)为
include_directories(${RKNN_SDK_PATH}/rknn_api/include)
#可能会报错，说thread_->Activate(-1); thread_->Deactivate(); 没有这个成员函数，注释掉它们即可。
8.测试
export LD_LIBRARY_PATH=/install:/install/lib
cp /rknn-api/Linux/tmp/* . -rf
../install/test_rknn_mobilenet 在./install/目录下有若干个程序可供测试使用。可以一一尝试。
谨记：如果更改了.cpp文件，一定要记得make && make install。