一、关键步骤

下载源码： ACL_YOLOV3

1.1 安装opencv

安装编译好的opencv静态链接库。

1.1.1 下载

链接：https://pan.baidu.com/s/1sdCsP3_NU48Nz8PYw-wpkg
提取码：1234

或者

链接: https://pan.baidu.com/s/1Z8sGoSI3eqUfime9aF5wSQ 密码: r80o

1.1.2 解压

解压opencv压缩包到指定目录，例如：/home/ma-user/work/opencv-4.5.1。

1.2 修改CMakeLists.txt

# 修改OpenCV_DIR路径
# OpenCV_DIR目录包含OpenCVConfig.cmake
set(OpenCV_DIR /home/ma-user/work/opencv-4.5.1/lib/cmake/opencv4)

# Find OpenCV, you may need to set OpenCV_DIR variable
# to the absolute path to the directory containing OpenCVConfig.cmake file
# via the command line or GUI
find_package(OpenCV REQUIRED)

# If the package has been found, several variables will
# be set, you can find the full list with descriptions
# in the OpenCVConfig.cmake file.
# Print some message showing some of them
if(${OPENCV_FOUND})
    message(STATUS "OpenCV version: ${OpenCV_VERSION}")
    message(STATUS "OpenCV include path: ${OpenCV_INCLUDE_DIRS}")
    message(STATUS "OpenCV libraries: ${OpenCV_LIBS}")
endif()

# Add OpenCV headers location to your include paths
include_directories(${OpenCV_INCLUDE_DIRS})

# 修改ACL_PATH路径
set(ACL_PATH /home/ma-user/Ascend/ascend-toolkit/latest)

1.3 模型转换

目标检测模型

1.3.1 模型下载

https://pan.baidu.com/s/1yiCrnmsOm0hbweJBiiUScQ

或者

链接：https://pan.baidu.com/s/1xZ2UuQAchtnpVrcCO6zQPA
提取码：1234

1.3.2 修改 prototxt 配置文件

完全按照模板修改：YOLOv3网络模型prototxt修改

1. upsample算子upsample_param属性参数修改；

   修改两处。

2. 新增三个Yolo算子；

3. 最后一层增加YoloV3DetectionOutput算子；

4. 新增输入；

   原输入：

   input: "data"
   input_shape {
         
       dim: 1
       dim: 3
       dim: 416
       dim: 416
   }
   

   新增输入：

   input: "img_info"
   input_shape {
         
     dim: 1
     dim: 4
   }
   

1.3.3 修改 aipp 配置文件

配置文件模板

不修改。

1.3.4 模型转换

cd yolov3
atc --model=./yolov3.prototxt \
    --weight=./yolov3.caffemodel \
    --framework=0 \
    --output=./yolov3_aipp \
    --output_type=FP32 \
    --soc_version=Ascend910 \
    --insert_op_conf=./aipp_yolov3.cfg

1.4 运行成功

[[email protected] acl_yolov3]$ ./ACL_YOLOV3 ./data/model/yolov3_aipp.om ./test.jpg
acl init successfully
Create context successfully
set context successfully
Create stream successfully
ModelProcess:Begin to init instance.
finish init AclProcess
ModelProcess:Begin to inference.
begin postprocess
The number of output buffers of yolov3 model is 2
Object detected number is 1
#Obj0, box(427.5, 424.25, 1513, 851.5) confidence: 1 lable: 2
model run success!
cost time : 18.924ms
ModelProcess:Begin to deinit instance.
ModelProcess:Finished deinit instance.
all tasks in stream done
Destroy Stream successfully
Destroy Context successfully
acl deinit successfully

二、FAQ

Q：缺少-fPIC

/usr/bin/ld: CMakeFiles/ACL_YOLOV3.dir/main.cpp.o: relocation R_AARCH64_ADR_PREL_PG_HI21 against symbol `[email protected]@GLIBCXX_3.4' which may bind externally can not be used when making a shared object; recompile with -fPIC /usr/bin/ld: CMakeFiles/ACL_YOLOV3.dir/main.cpp.o(.text+0x2c): unresolvable R_AARCH64_ADR_PREL_PG_HI21 relocation against symbol `[email protected]@GLIBCXX_3.4'
/usr/bin/ld: final link failed: bad value
collect2: error: ld returned 1 exit status
make[2]: *** [CMakeFiles/ACL_YOLOV3.dir/build.make:144: ../ACL_YOLOV3] Error 1
make[1]: *** [CMakeFiles/Makefile2:76: CMakeFiles/ACL_YOLOV3.dir/all] Error 2
make: *** [Makefile:84: all] Error 2

解决办法：
修改CMakeLists.txt
add_definitions(-fPIC)

Q：缺少 dlsym

/usr/bin/ld: /home/ma-user/work/opencv-4.5.1/lib/libopencv_core.a(filesystem.cpp.obj): in function `cv::plugin::impl::DynamicLib::getSymbol(char const*) const': filesystem.cpp:(.text._ZNK2cv6plugin4impl10DynamicLib9getSymbolEPKc+0x8): undefined reference to `dlsym' /usr/bin/ld: /home/ma-user/work/opencv-4.5.1/lib/libopencv_core.a(filesystem.cpp.obj): in function `cv::plugin::impl::DynamicLib::libraryLoad(std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&)':
filesystem.cpp:(.text._ZN2cv6plugin4impl10DynamicLib11libraryLoadERKNSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEE+0x3c): undefined reference to `dlopen' /usr/bin/ld: /home/ma-user/work/opencv-4.5.1/lib/libopencv_core.a(filesystem.cpp.obj): in function `cv::plugin::impl::DynamicLib::libraryRelease()': filesystem.cpp:(.text._ZN2cv6plugin4impl10DynamicLib14libraryReleaseEv+0x3b4): undefined reference to `dlclose'
/usr/bin/ld: /home/ma-user/work/opencv-4.5.1/lib/libopencv_core.a(datafile.cpp.obj): in function `cv::utils::getModuleLocation(void const*)': datafile.cpp:(.text._ZN2cv5utilsL17getModuleLocationEPKv+0x2c): undefined reference to `dladdr' /usr/bin/ld: /home/ma-user/work/opencv-4.5.1/lib/libopencv_core.a(datafile.cpp.obj): in function `cv::utils::getBinLocation(std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >&)':
datafile.cpp:(.text._ZN2cv5utils14getBinLocationERNSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEE+0x30): undefined reference to `dladdr'
collect2: error: ld returned 1 exit status
make[2]: *** [CMakeFiles/ACL_YOLOV3.dir/build.make:144: ../ACL_YOLOV3] Error 1
make[1]: *** [CMakeFiles/Makefile2:76: CMakeFiles/ACL_YOLOV3.dir/all] Error 2
make: *** [Makefile:84: all] Error 2

解决办法：
修改CMakeLists.txt，添加dl
target_link_libraries(${PROJECT_NAME} ascendcl ${OpenCV_LIBS} pthread -Wl,-z,relro,-z,now,-z,noexecstack -pie -s stdc++)
修改为
target_link_libraries(${PROJECT_NAME} ascendcl ${OpenCV_LIBS} pthread -Wl,-z,relro,-z,now,-z,noexecstack -pie -s stdc++ dl)

Q：缺少acl.json文件

[[email protected] acl_yolov3]$ ./ACL_YOLOV3 ./data/model/yolov3_aipp.om ./test.jpg
Failed to init acl, ret = 100003
AclProcess Init faild.
some tasks in stream not done, ret = 107002
all tasks in stream done
Destroy Stream faild, ret = 100000
Destroy Stream successfully
Destroy Context faild, ret = 100000
Destroy Context successfully
acl deinit successfully

错误原因：
缺少acl.json文件，导致aclInit初始化失败

解决办法：
aclError ret = aclprocess.Init(0,"./data/acl.json",argv[1]);
改为
aclError ret = aclprocess.Init(0,"",argv[1]);

Q：保存图片无法显示

错误原因：
由于采用opencv静态链接库，导致jpg编码器无法使用（功能不完整）

解决办法：
使用png格式保存图片

imwrite("result.jpg",img);
改为
imwrite("result.png",img);