ubuntu18.04+OpenCV4.2+Darknet2NCNN将yolov3模型转换为Android-NCNN模型

浅浅的花香味﹌ 2023-10-07 11:23 18阅读 0赞

## ubuntu18.04+OpenCV4.2+Darknet2NCNN将yolov3模型转换为ncnn模型 ##

opencv版本3.x可能更好一些，3到4有些参数变化

一、下载[darknet2ncnn][]

可以直接去官网下载，如果配置了gitee，也可以用命令

git clone https://github.com/xiangweizeng/darknet2ncnn.git

二、编译电脑端模型并测试

1.根据官网教程来，初始化，我未配置git所以跳过了：

cd darknet2ncnn
    git submodule init
    git submodule update

2.darknet2ncnn解压后，根目录下有darknet、ncnn两个文件夹，里面是空的，需要自己下载[darknet][]和[ncnn][]放进去

3.构建 darknet

cd darknet2
    make -j8
    rm libdarknet.so

4.构建 ncnn

# workspace darknet2ncnn
    cd ncnn
    mkdir build
    cd build
    cmake ..
    make -j8
    make install
    cd ../../

5.构建 darknet2ncnn , convert\_verify and libdarknet2ncnn.a

# workspace darknet2ncnn
    make -j8

6.yolov3-tiny模型转换及验证

make yolov3-tiny.net

转换结果，ncnn模型需要的网络和权重文件的格式：

./darknet2ncnn data/yolov3-tiny.cfg  data/yolov3-tiny.weights example/zoo/yolov3-tiny.param  example/zoo/yolov3-tiny.bin

7.构建 example

#workspace darknet2ncnn
    cd example
    make -j4

注意，可能会报错：

A：因为CV\_LOAD\_IMAGE\_COLOR在OpenCV4版本命名为：IMREAD\_COLOR  
修改方法：将错误参数替换为：cv::IMREAD\_COLOR  
![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MjE3NjYzOQ_size_16_color_FFFFFF_t_70]  
B：CV\_FILLED，这个参数可以控制输出图片识别结果标识的边框，未找到应该替换成什么枚举值，所以我直接改成参数：1 了

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MjE3NjYzOQ_size_16_color_FFFFFF_t_70 1]8.运行yolov3-tiny

#workspace example
    make yolov3-tiny.coco

9.最终结果如下：

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MjE3NjYzOQ_size_16_color_FFFFFF_t_70 2]  
![在这里插入图片描述][20200320002814652.png]

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MjE3NjYzOQ_size_16_color_FFFFFF_t_70 3]  
三、编译Android端NCNN库文件，前提条件：已经配置了Android的SDK

[官网编译教程][Link 1]

1.进入NCNN执行：`mkdir -p build-android-arm64-v8a`

根据你自己调试用的手机去选择编译的版本：‘arm64-v8a’,‘armeabi-v7a’,‘x86’,‘x86\_64’

2.进入你新建的目录：`cd build-android-arm64-v8a`

3.编译：  
`cmake -DCMAKE_TOOLCHAIN_FILE=/home/zhx/Android/Sdk/ndk/21.0.6113669/build/cmake/android.toolchain.cmake \ -DANDROID_ABI="arm64-v8a" \ -DANDROID_ARM_NEON=ON ..`

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MjE3NjYzOQ_size_16_color_FFFFFF_t_70 4]4.编译：

`make -j4`

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MjE3NjYzOQ_size_16_color_FFFFFF_t_70 5]  
`make install`

![在这里插入图片描述][watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MjE3NjYzOQ_size_16_color_FFFFFF_t_70 6]  
在build-android-arm64-v8a文件夹下生成了install文件夹，它就是移植到Android端需要的库文件

[darknet2ncnn]: https://gitee.com/damone/darknet2ncnn?_from=gitee_search
[darknet]: https://pjreddie.com/darknet/yolo/
[ncnn]: https://github.com/Tencent/ncnn
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MjE3NjYzOQ_size_16_color_FFFFFF_t_70]: https://img-blog.csdnimg.cn/20200320001042947.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MjE3NjYzOQ==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MjE3NjYzOQ_size_16_color_FFFFFF_t_70 1]: https://img-blog.csdnimg.cn/20200320001526558.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MjE3NjYzOQ==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MjE3NjYzOQ_size_16_color_FFFFFF_t_70 2]: https://img-blog.csdnimg.cn/20200320002801227.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MjE3NjYzOQ==,size_16,color_FFFFFF,t_70
[20200320002814652.png]: https://img-blog.csdnimg.cn/20200320002814652.png
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MjE3NjYzOQ_size_16_color_FFFFFF_t_70 3]: https://img-blog.csdnimg.cn/20200320002831163.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MjE3NjYzOQ==,size_16,color_FFFFFF,t_70
[Link 1]: https://github.com/Tencent/ncnn/wiki/how-to-build#build-for-android
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MjE3NjYzOQ_size_16_color_FFFFFF_t_70 4]: https://img-blog.csdnimg.cn/20200321180706715.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MjE3NjYzOQ==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MjE3NjYzOQ_size_16_color_FFFFFF_t_70 5]: https://img-blog.csdnimg.cn/20200321183131529.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MjE3NjYzOQ==,size_16,color_FFFFFF,t_70
[watermark_type_ZmFuZ3poZW5naGVpdGk_shadow_10_text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MjE3NjYzOQ_size_16_color_FFFFFF_t_70 6]: https://img-blog.csdnimg.cn/20200321183649749.png?x-oss-process=image/watermark,type_ZmFuZ3poZW5naGVpdGk,shadow_10,text_aHR0cHM6Ly9ibG9nLmNzZG4ubmV0L3dlaXhpbl80MjE3NjYzOQ==,size_16,color_FFFFFF,t_70