RK1126 NPU yolov5 6.2

基于 rk npu ， 实现 yolov5 6.2 模型推理

---

实现过程 ⚡️​

编译 opencv

1. 需根据自己路径修改.

	cmake -D CMAKE_BUILD_TYPE=RELEASE \
	      -D CMAKE_C_COMPILER=./gcc-arm-8.3-2019.02-x86_64-arm-linux-gnueabihf/bin/arm-linux-gnueabihf-gcc \
	      -D CMAKE_CXX_COMPILER=./gcc-arm-8.3-2019.02-x86_64-arm-linux-gnueabihf/bin/arm-linux-gnueabihf-g++ \
	      -D CMAKE_INSTALL_PREFIX=/usr/local/opencv4.2_gnueabihf \
	      -D CMAKE_TOOLCHAIN_FILE=../platforms/linux/arm-gnueabi.toolchain.cmake \
	      -D OPENCV_FORCE_3RDPARTY_BUILD=ON \
	      -D BUILD_ZLIB=ON \
	      -D INSTALL_C_EXAMPLES=OFF \
	      -D INSTALL_PYTHON_EXAMPLES=OFF \
	      -D WITH_TBB=OFF \
	      -D WITH_V4L=ON \
	      -D WITH_CUBLAS=1 \
	      -D OPENCV_SKIP_PYTHON_LOADER=ON \
	      -D OPENCV_GENERATE_PKGCONFIG=ON \
	      -D WITH_QT=OFF \
	      -D WITH_OPENGL=ON \
	      -D WITH_FFMPEG=ON \
	      -D OPENCV_EXTRA_MODULES_PATH=../../opencv_contrib-4.2.0/modules \
	      -D ENABLE_CXX11=ON \
	      -D BUILD_EXAMPLES=OFF .. \
	      -D BUILD_opencv_world=ON ..
	  make -j4
	  sudo make install

Q&A ❓​

1. zlib.h：没有那个文件或目录 #include

   -D OPENCV_FORCE_3RDPARTY_BUILD=ON \
   -D BUILD_ZLIB=ON \
   

2. opencv G_STATIC_ASSERT(sizeof (unsigned long long) == sizeof (guint64))

   -D CMAKE_TOOLCHAIN_FILE=../platforms/linux/arm-gnueabi.toolchain.cmake
   

Demo

1. 参考: https://github.com/shaoshengsong/rockchip_rknn_yolov5
2. 这里修改了下前处理,使用 letterbox

   cv::Mat letterbox(const cv::Mat &img, cv::Size new_shape, cv::Scalar color, bool _auto, bool scaleFill, bool scaleup, int stride)
   {
       float width = img.cols;
       float height = img.rows;
       float r = cv::min(new_shape.width / width, new_shape.height / height);
       if (!scaleup)
           r = cv::min(r, 1.0f);
       int new_unpadW = int(round(width * r));
       int new_unpadH = int(round(height * r));
       int dw = new_shape.width - new_unpadW;
       int dh = new_shape.height - new_unpadH;
       if (_auto)
       {
           dw %= stride;
           dh %= stride;
       }
       dw /= 2, dh /= 2;
       cv::Mat dst;
       resize(img, dst, cv::Size(new_unpadW, new_unpadH), 0, 0, cv::INTER_LINEAR);
       int top = int(round(dh - 0.1));
       int bottom = int(round(dh + 0.1));
       int left = int(round(dw - 0.1));
       int right = int(round(dw + 0.1));
       copyMakeBorder(dst, dst, top, bottom, left, right, cv::BORDER_CONSTANT, color);
   
       pad_left_ = left;
       pad_top_ = top;
       scale_ = r;
       return dst;
   }
   

3. inference

   int inference(const cv::Mat &orig_img, rt_detect_result_group_t &detect_result_group){
   	...
   	cv::Mat img = letterbox(orig_img, cv::Size(width_, height_), cv::Scalar(114, 114, 114), false, false, false, 32);
   	cv::Mat resimg;
   	// cv::resize(orig_img, resimg, cv::Size(width_, height_), (0, 0), (0, 0), cv::INTER_LINEAR);
   	cv::cvtColor(img, resimg, cv::COLOR_BGR2RGB);
    ...
   }
   

4. 执行 ./build.sh
5. 我用的 adb ，将程序拷贝到 rk1126 板子上.

   adb.exe push xxx  xxxx
   

---

参考文章 ​

1. https://github.com/shaoshengsong/rockchip_rknn_yolov5
2. https://github.com/rockchip-linux/rknpu