Linux编译(cross-compilation)
主机host系统：ubuntu16.04
目标硬件：ARM imx6qsabresd CortexA9
交叉编译可执行文件格式：ELF 32-bit LSB executable, ARM, EABI5 version 1 (SYSV), dynamically linked, interpreter /lib/ld-linux-armhf.so.3, for GNU/Linux 3.2.0
1.交叉编译
1.1 安装编译环境
sudo apt-get install gcc-arm-linux-gnueabihf g++-arm-linux-gnueabihf g++-multilib-arm-linux-gnueabihf pkg-config-arm-linux-gnueabihf sudo apt-get install gcc-arm-linux-gnueabihf g++-arm-linux-gnueabihf g++-multilib-arm-linux-gnueabihf pkg-config-arm-linux-gnueabihf 完全不需要虚拟机编译出一个与ARM板子相同的交叉编译环境，以上环境能兼容。
1.2 交叉编译helloworld
#include
int main()
{
std::cout<<"hello world!"< arm-linux-gnueabihf-g++ helloworld.cpp -o helloworld通过以下命令查看兼容性（是否能在ARM板子上执行）
file helloworldfile helloworld最后上传或者拷贝到ARM板子中执行验证即可。
1.3 交叉编译opencv
参考网站：https://docs.opencv.org/3.4.3/d0/d76/tutorial_arm_crosscompile_with_cmake.html
修改opencv/platforms/linux目录下的arm-gnueabi.toolchain.cmake中的CMAKE_CXX_COMPILER（arm-linux-gnueabihf-g++），以及CMAKE_CXX_FLAGS（可能用到-march=armv7-a -mfpu=neon -mfloat-abi=hard -mcpu=cortex-a9），CMAKE_EXE_LINKER_FLAGS（opencv需要的-lopencv_core -lopencv_imgproc -lopencv_imgcodecs），示例：
set(CMAKE_SYSTEM_NAME Linux)
#set( CMAKE_SYSTEM_PROCESSOR arm )
set( CMAKE_C_COMPILER arm-linux-gnueabihf-gcc)
set( CMAKE_CXX_COMPILER arm-linux-gnueabihf-g++)
SET(CMAKE_CXX_FLAGS "-march=armv7-a -mfpu=neon -mfloat-abi=hard -mcpu=cortex-a9 -c -Wall ")
set(CMAKE_EXE_LINKER_FLAGS "-lopencv_core -lopencv_imgproc -lopencv_imgcodecs")
SET(CMAKE_FIND_ROOT_PATH_MODE_PROGRAM NEVER)
# for libraries and headers in the target directories
SET(CMAKE_FIND_ROOT_PATH_MODE_LIBRARY ONLY)
SET(CMAKE_FIND_ROOT_PATH_MODE_INCLUDE ONLY)执行：
cmake -DCMAKE_TOOLCHAIN_FILE=../arm-gnueabi.toolchain.cmake ../../..cmake -DCMAKE_TOOLCHAIN_FILE=../arm-gnueabi.toolchain.cmake ../../..然后make即可
- 编译安装到指定路径(目录)
如果需要编译(bin/host/lib/shared)到指定目录(如host)下，添加参数 -D CMAKE_INSTALL_PREFIX=./host即可，然后执行make install，完成之后即可在build/host目录下找到(bin/host/lib/shared)文件。
#opencv目录下自建build目录，并切换到build下
cmake -D CMAKE_INSTALL_PREFIX=./host ..
make install -j8find_package(OpenCV REQUIRED)找不到OpenCV时，CMakeLists中添加自定义安装的opencv库，参考：自定义引入OpenCV版本
2.编译动态链接.so(cross-compile)
参考网站：https://www.cnblogs.com/52php/p/5681711.html
//....test.c....
#include
int fun()
{
return 1;
} arm-linux-gnueabihf-gcc test.c -I./ -fPIC -shared -o libtest.so编译可执行文件若需要调用libtest.so文件，直接采用"-ltest"即可链接到libtest.so文件。
3.交叉编译人脸检测(Shiguang Shan)
在SeetaFaceEngine-master/FaceDetection目录下，修改CMakeLists.txt
- 修改arm兼容的交叉编译环境
- 各个库修改为arm兼容的库(最重要的是修改arm的opencv头文件以及库文件)
- OPENMP以及SSE设置为OFF
- EXAMPLES根据自己的需要决定
cmake_minimum_required(VERSION 3.1.0)
project(seeta_facedet_lib)
set(CMAKE_SYSTEM_NAME Linux)
#set( CMAKE_SYSTEM_PROCESSOR arm )
set( CMAKE_C_COMPILER arm-linux-gnueabihf-gcc)
set( CMAKE_CXX_COMPILER arm-linux-gnueabihf-g++)
SET(CMAKE_CXX_FLAGS "-march=armv7-a -mfpu=neon -mfloat-abi=hard -mcpu=cortex-a9")
#set(CMAKE_EXE_LINKER_FLAGS "-lopencv_core -lopencv_imgproc -lopencv_imgcodecs")
SET(CMAKE_FIND_ROOT_PATH_MODE_PROGRAM NEVER)
# for libraries and headers in the target directories
SET(CMAKE_FIND_ROOT_PATH_MODE_LIBRARY ONLY)
SET(CMAKE_FIND_ROOT_PATH_MODE_INCLUDE ONLY)
set(CMAKE_BUILD_TYPE Debug)
# Build options
option(BUILD_EXAMPLES "Set to ON to build examples" ON)
option(USE_OPENMP "Set to ON to build use openmp" OFF)
option(USE_SSE "Set to ON to build use SSE" OFF)
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -O2")
# Use C++11
set(CMAKE_CXX_STANDARD 11)
set(CMAKE_CXX_STANDARD_REQUIRED ON)
message(STATUS "C++11 support has been enabled by default.")
# Use SSE
if (USE_SSE)
add_definitions(-DUSE_SSE)
message(STATUS "Use SSE")
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -msse4.1")
endif()
# Use OpenMP
if (USE_OPENMP)
find_package(OpenMP QUIET)
if (OPENMP_FOUND)
message(STATUS "Use OpenMP")
add_definitions(-DUSE_OPENMP)
set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${OpenMP_C_FLAGS}")
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${OpenMP_CXX_FLAGS}")
set(CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS} ${OpenMP_EXE_LINKER_FLAGS}")
endif()
endif()
include_directories(include)
set(src_files
src/util/nms.cpp
src/util/image_pyramid.cpp
src/io/lab_boost_model_reader.cpp
src/io/surf_mlp_model_reader.cpp
src/feat/lab_feature_map.cpp
src/feat/surf_feature_map.cpp
src/classifier/lab_boosted_classifier.cpp
src/classifier/mlp.cpp
src/classifier/surf_mlp.cpp
src/face_detection.cpp
src/fust.cpp
)
# Build shared library
add_library(seeta_facedet_lib SHARED ${src_files})
set(facedet_required_libs seeta_facedet_lib)
# Build examples
if (BUILD_EXAMPLES)
message(STATUS "Build with examples.")
#find_package(OpenCV)
include_directories("../../../OpenCV_ARM/include")
# if (NOT OpenCV_FOUND)
# message(WARNING "OpenCV not found. Test will not be built.")
# else()
# include_directories(${OpenCV_INCLUDE_DIRS})
# list(APPEND facedet_required_libs ${OpenCV_LIBS})
add_executable(facedet_test src/test/facedetection_test.cpp)
target_link_libraries(facedet_test ${facedet_required_libs} -lopencv_core -lopencv_imgproc -lopencv_imgcodecs -lopencv_videoio -lopencv_video -L../../../OpenCV_ARM/lib)
# endif()
endif()
4.CMake基础入门
参考网站：https://cmake.org/cmake/help/v3.14/manual/cmake.1.html
- 基础认识：
CMake是用来将一个项目源代码编译成一个项目的可执行文件。
- 使用：
mkdir build
cd build
cmake ../ #编译CMakeLists.txt中写好的内容，产生CMakeCache.txt
makemkdir build
cd build
cmake ../ #编译CMakeLists.txt中写好的内容，产生CMakeCache.txt
makeCMakeLists.txt文件规定了编译目标，以及项目的依赖。
CMakeCache.txt在执行cmake命令后产生，用一个树状结构标识项目目录，存储固定的信息。
- CMakeLists.txt编写
- 步骤：
示例参考：https://www.cnblogs.com/cv-pr/p/6206921.html
①指定cmake版本
cmake_minimum_required(VERSION 3.2)②指定项目名称，一般与项目文件夹名相同即可
PROJECT(Test)③添加头文件目录（它相当于g++选项中的-I参数的作用，也相当于环境变量中增加路径到CPLUS_INCLUDE_PATH变量的作用）
INCLUDE_DIRECTORIES(
include
)④添加源文件目录
AUX_SOURCE_DIRECTORY(src DIR_SRCS)⑤设置环境变量，编译用到的源文件全部需要添加
SET(TEST_MATH
${DIR_SRCS}
)⑥添加可执行文件
ADD_EXECUTABLE(${PROJECT_NAME} ${TEST})⑦添加可执行文件所需库文件
TARGET_LINK_LIBRARIES(${PROJECT_NAME} -lopencv_core)多种写法：
#比如（以下写法（包括备注中的）都可以）：
TARGET_LINK_LIBRARIES(myProject hello) # 连接libhello.so库
TARGET_LINK_LIBRARIES(myProject libhello.a)
TARGET_LINK_LIBRARIES(myProject libhello.so)
#再如：
TARGET_LINK_LIBRARIES(myProject libeng.so) #这些库名写法都可以。
TARGET_LINK_LIBRARIES(myProject eng)
TARGET_LINK_LIBRARIES(myProject -leng)
以上cmake基本步骤算是完成，还可以添加其他的功能：
LINK_DIRECTORIES（添加需要链接的库文件目录）
link_directories(directory1 directory2 ...) #它相当于g++命令的-L选项的作用，也相当于环境变量中增加LD_LIBRARY_PATH的路径的作用LINK_LIBRARIES （添加需要链接的库文件路径，注意这里是全路径。还是推荐使用⑦）
LINK_LIBRARIES("/opt/MATLAB/R2012a/bin/glnxa64/libeng.so")add_subdirectory(NAME) (添加一个文件夹进行编译，该文件夹下的CMakeLists.txt 负责编译该文件夹下的源码. NAME是相对于调用add_subdirectory的CMakeListst.txt的相对路径)
5.交叉编译人脸检测+人脸对齐(Shiguang Shan)
step1:SeetaFaceEngine-master目录下创建build文件夹，以及CMakeLists.txt(内容如下)
step2:build目录下编译
cmake_minimum_required(VERSION 2.8.4)
project(seeta_fa_lib)
set(CMAKE_SYSTEM_NAME Linux)
#set( CMAKE_SYSTEM_PROCESSOR arm )
set( CMAKE_C_COMPILER arm-linux-gnueabihf-gcc)
set( CMAKE_CXX_COMPILER arm-linux-gnueabihf-g++)
SET(CMAKE_CXX_FLAGS "-march=armv7-a -mfpu=neon -mfloat-abi=hard -mcpu=cortex-a9")
SET(CMAKE_FIND_ROOT_PATH_MODE_PROGRAM NEVER)
# for libraries and headers in the target directories
SET(CMAKE_FIND_ROOT_PATH_MODE_LIBRARY ONLY)
SET(CMAKE_FIND_ROOT_PATH_MODE_INCLUDE ONLY)
# Build options
option(BUILD_EXAMPLES "Set to ON to build examples" ON)
# Use C++11
set(CMAKE_CXX_STANDARD 11)
set(CMAKE_CXX_STANDARD_REQUIRED ON)
message(STATUS "C++11 support has been enabled by default.")
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -O2")
set(fd FaceDetection) #起别名，后面引用带${...}
set(fa FaceAlignment)
include_directories(${fd}/include)
include_directories(${fa}/include)
set(fd_src_files
${fd}/src/util/nms.cpp
${fd}/src/util/image_pyramid.cpp
${fd}/src/io/lab_boost_model_reader.cpp
${fd}/src/io/surf_mlp_model_reader.cpp
${fd}/src/feat/lab_feature_map.cpp
${fd}/src/feat/surf_feature_map.cpp
${fd}/src/classifier/lab_boosted_classifier.cpp
${fd}/src/classifier/mlp.cpp
${fd}/src/classifier/surf_mlp.cpp
${fd}/src/face_detection.cpp
${fd}/src/fust.cpp
)
add_library(seeta_facedet_lib SHARED ${fd_src_files})
set(facedet_required_libs seeta_facedet_lib)
set(fa_src_files
${fa}/src/cfan.cpp
${fa}/src/face_alignment.cpp
${fa}/src/sift.cpp
)
add_library(seeta_fa_lib SHARED ${fa_src_files})
set(fa_required_libs seeta_fa_lib)
if (BUILD_EXAMPLES)
message(STATUS "Build with examples.")
include_directories("../../OpenCV_ARM/include")
include_directories("../../OpenCV_ARM/include/opencv")
include_directories("../../OpenCV_ARM/include/opencv2")
link_directories("../../OpenCV_ARM/lib")
add_executable(fa_test ${fa}/src/test/face_alignment_test.cpp)
target_link_libraries(fa_test ${fa_required_libs} ${facedet_required_libs} -lopencv_core -lopencv_imgproc -lopencv_imgcodecs -lopencv_videoio -lopencv_video -L../../OpenCV_ARM/lib)
endif()imx6q运行时间：
real 0m3.526s
user 0m3.410s
sys 0m0.110s
6.交叉编译人脸检测+人脸对齐+人脸验证(Shiguang Shan)
主要遇到的问题：
- 采用add_subdirectories()方法，找不到自己编译好的OpenCV_ARM中的头文件(include)和库文件(lib)，因此我将所有的CMakeLists中的功能写到一个中；
- FaceIdentification/src/test/test_face_verification.cpp中包含的OpenCV头文件可能存在路径问题，比如我修改了#include "opencv2/highgui/highgui.hpp"
#基本配置环境
cmake_minimum_required(VERSION 2.8.4)
project(seeta_fa_lib)
set(CMAKE_SYSTEM_NAME Linux)
#set( CMAKE_SYSTEM_PROCESSOR arm )
set( CMAKE_C_COMPILER arm-linux-gnueabihf-gcc)
set( CMAKE_CXX_COMPILER arm-linux-gnueabihf-g++)
SET(CMAKE_CXX_FLAGS "-march=armv7-a -mfpu=neon -mfloat-abi=hard -mcpu=cortex-a9")
SET(CMAKE_FIND_ROOT_PATH_MODE_PROGRAM NEVER)
# for libraries and headers in the target directories
SET(CMAKE_FIND_ROOT_PATH_MODE_LIBRARY ONLY)
SET(CMAKE_FIND_ROOT_PATH_MODE_INCLUDE ONLY)
# Build options
option(BUILD_EXAMPLES "Set to ON to build examples" ON)
# Use C++11
set(CMAKE_CXX_STANDARD 11)
set(CMAKE_CXX_STANDARD_REQUIRED ON)
message(STATUS "C++11 support has been enabled by default.")
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -O2")
#人脸检测(FaceDetection)+人脸对齐(FaceAlignment)
set(fd FaceDetection)
set(fa FaceAlignment)
include_directories(${fd}/include)
include_directories(${fa}/include)
set(fd_src_files
${fd}/src/util/${fd}/src/classifier/surf_mlp.cpp
${fd}/src/face_detection.cpp
${fd}/src/fust.cpp
)
add_library(seeta_facedet_lib SHARED ${fd_src_files})
set(facedet_required_libs seeta_facedet_lib)
set(fa_src_files
${fa}/src/cfan.cpp
${fa}/src/face_alignment.cpp
${fa}/src/sift.cpp
)
add_library(seeta_fa_lib SHARED ${fa_src_files})
set(fa_required_libs seeta_fa_lib)
#人脸识别(FaceIdentification)的配置环境
set (VIPLNET_VERSION_MAJOR 4)
set (VIPLNET_VERSION_MINOR 5)
set(CMAKE_BUILD_TYPE "Release")
set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS} -std=c++11 -O2 -g -ggdb")
set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS} -std=c++11 -O2")
MESSAGE(STATUS "other platform: ${CMAKE_SYSTEM_NAME}")
set(fi FaceIdentification)
set(VIPLNET_INCLUDE_DIR ${fi}/include)
set(VIPLNET_SRC_DIR ${fi}/src)
# set __VIOL_LOG__ macro
# add_definitions(-D__VIPL_LOG__)
include_directories(${VIPLNET_INCLUDE_DIR})
include_directories(${VIPLNET_SRC_DIR})
#add_subdirectory(${fi}/src)
aux_source_directory(${fi}/src SRC_LIST)
aux_source_directory(${fi}/tools TOOLS_LIST)
${fd}/src/util/image_pyramid.cpp
${fd}/src/io/lab_boost_model_reader.cpp
${fd}/src/io/surf_mlp_model_reader.cpp
${fd}/src/feat/lab_feature_map.cpp
${fd}/src/feat/surf_feature_map.cpp
${fd}/src/classifier/lab_boosted_classifier.cpp
${fd}/src/classifier/mlp.cpp
${fd}/src/classifier/surf_mlp.cpp
${fd}/src/face_detection.cpp
${fd}/src/fust.cpp
)
add_library(seeta_facedet_lib SHARED ${fd_src_files})
set(facedet_required_libs seeta_facedet_lib)
set(fa_src_files
${fa}/src/cfan.cpp
${fa}/src/face_alignment.cpp
${fa}/src/sift.cpp
)
add_library(seeta_fa_lib SHARED ${fa_src_files})
set(fa_required_libs seeta_fa_lib)
#FaceIdentification
set (VIPLNET_VERSION_MAJOR 4)
set (VIPLNET_VERSION_MINOR 5)
set(CMAKE_BUILD_TYPE "Release")
set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS} -std=c++11 -O2 -g -ggdb")
set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS} -std=c++11 -O2")
MESSAGE(STATUS "other platform: ${CMAKE_SYSTEM_NAME}")
set(fi FaceIdentification)
set(VIPLNET_INCLUDE_DIR ${fi}/include)
set(VIPLNET_SRC_DIR ${fi}/src)
# set __VIOL_LOG__ macro
# add_definitions(-D__VIPL_LOG__)
include_directories(${VIPLNET_INCLUDE_DIR})
include_directories(${VIPLNET_SRC_DIR})
#add_subdirectory(${fi}/src)
aux_source_directory(${fi}/src SRC_LIST)
aux_source_directory(${fi}/tools TOOLS_LIST)
add_library(viplnet SHARED ${SRC_LIST} ${TOOLS_LIST})
set_target_properties(viplnet PROPERTIES
VERSION ${VIPLNET_VERSION_MAJOR}.${VIPLNET_VERSION_MINOR}
SOVERSION ${VIPLNET_VERSION_MAJOR}.${VIPLNET_VERSION_MINOR})
if (BUILD_EXAMPLES)
message(STATUS "Build with examples.")
include_directories("../../OpenCV_ARM/include")
include_directories("../../OpenCV_ARM/include/opencv")
include_directories("../../OpenCV_ARM/include/opencv2")
link_directories("../../OpenCV_ARM/lib")
add_executable(fi_verification_test ${fi}/src/test/test_face_verification.cpp)
target_link_libraries(fi_verification_test ${fa_required_libs} ${facedet_required_libs} viplnet -lopencv_core -lopencv_imgproc -lopencv_imgcodecs -lopencv_videoio -lopencv_video -L../../OpenCV_ARM/lib)
endif()计算同一张图片的识别效果，即计算同一张图片的相似度（识别率=1），计算同一人不同图片（本人一张正装照和手机拍的照片，识别率=0.643433）。因此，不同设备下，识别精度不会损失，识别时间在arm平台下会超级慢，识别时间如下：
| Freescale i.MX6 Quad/DualLite (Device Tree)运行时间 | Intel(R) Core(TM) i7-7800X CPU @ 3.50GHz运行时间 |
| real 0m13.799s | real 0m1.007s |
| user 0m12.870s | user 0m1.638s |
| sys 0m0.970s | sys 0m0.330s |
7.树莓派( cross-compilation )
树莓派连接两个USB摄像头，同时获取2个摄像头的frame，保存为图片。（两个摄像头不能接在一个usb hub上）
#include
#include "unistd.h"
#include "opencv2/imgproc.hpp"
#include "opencv2/highgui.hpp"
#include "opencv2/videoio.hpp"
using namespace std;
int main(int argc, char *argv[])
{
cout<<"this is a test program!"<>frame0;
if(!frame0.empty())
{
cv::imwrite("0_1.jpg",frame0);
cout<<"0_1.jpg"<>frame1;
if(!frame1.empty())
{
cv::imwrite("1_1.jpg",frame1);
cout<<"1_1.jpg"< 树莓派ARMv7l实现时间延时：使用unistd.h中的sleep()即可。
交叉编译CMakeLists.txt文件如下：
cmake_minimum_required(VERSION 2.8)
cmake_policy(SET CMP0015 NEW)
project(opencv_camera)
set(CMAKE_SYSTEM_NAME Linux)
set( CMAKE_C_COMPILER arm-linux-gnueabihf-gcc)
set( CMAKE_CXX_COMPILER arm-linux-gnueabihf-g++)
SET(CMAKE_FIND_ROOT_PATH_MODE_PROGRAM NEVER)
# for libraries and headers in the target directories
SET(CMAKE_FIND_ROOT_PATH_MODE_LIBRARY ONLY)
SET(CMAKE_FIND_ROOT_PATH_MODE_INCLUDE ONLY)
SET(CMAKE_CXX_FLAGS "-march=armv7-a -mfpu=neon -mfloat-abi=hard")
set(CMAKE_BUILD_TYPE Debug)
# Use C++11
set(CMAKE_CXX_STANDARD 11)
set(CMAKE_CXX_STANDARD_REQUIRED ON)
message(STATUS "C++11 support has been enabled by default.")
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -O2")
include_directories(
include
include/opencv
include/opencv2
)
link_directories(lib)
set(src opencv_camera.cpp)
add_executable(opencv_camera ${src})
target_link_libraries(opencv_camera libopencv_core.so libopencv_imgproc.so libopencv_imgcodecs.so libopencv_videoio.so libopencv_video.so libopencv_highgui.so)将前文交叉编译完成的OpenCV库文件lib，与OpenCV会用到的头文件include复制到CMakeLists.txt所在目录下。
注意：复制lib和include之后，如果编译可执行文件时，还是存在找不到OpenCV相关的库，或者OpenCV相关文件未定义的错误，则需考虑include_directories以及link_directories中包含的目录不准确，或者存在漏库/头文件的情况。
8.librealsense编译
测试系统：ubuntu16.04
测试硬件：Intel RealSense D435
8.1 编译SDK
- 新建一个项目目录intelrealsense
- 下载
终端切换至intelrealsense目录，git clone https://github.com/IntelRealSense/librealsense.git
- 编译
参考网站：https://dev.intelrealsense.com/docs/compiling-librealsense-for-linux-ubuntu-guide
如果需要编译在arm64(rockchip rk3399 aarch64)上，参考网站编译，则直接在rockpi终端，librealsense目录下运行./scripts/libuvc_installation.sh即可，当然也可以加入一些优化选项，比如将scripts/libuvc_installation.sh中cmake ../ -DFORCE_LIBUVC=true -DCMAKE_BUILD_TYPE=release修改为：cmake ../ -DFORCE_LIBUVC=true -DCMAKE_BUILD_TYPE=release -DBUILD_EXAMPLES=OFF -DBUILD_GRAPHICAL_EXAMPLES=OFF -DBUILD_WITH_OPENMP=OFF -DBUILD_UNIT_TESTS=OFF -DENABLE_ZERO_COPY=ON
8.2 编译测距示例
参考网站：https://blog.csdn.net/dieju8330/article/details/85420584
- 网站中的测距程序和CMakeLists.txt放于intelrealsense下，并在intelrealsense目录下新建一个lib目录；
- 将8.1中编译生成的librealsense2.so.2.21.0复制到lib目录下；
- CMakeLists.txt修改为
project(measure_distance)
cmake_minimum_required(VERSION 2.8)
cmake_policy(SET CMP0015 NEW)
include_directories("librealsense/include/")
set(CMAKE_CXX_FLAGS "-std=c++11")
#寻找opencv库
find_package(OpenCV REQUIRED)
#message(STATUS ${OpenCV_INCLUDE_DIRS})
#添加头文件
include_directories(${OpenCV_INCLUDE_DIRS})
link_directories(lib) #添加复制出来的链接库
add_executable(measure_distance measure_distance.cpp)
#链接Opencv库
target_link_libraries(measure_distance ${OpenCV_LIBS} )
#添加后可进行调试
set( CMAKE_BUILD_TYPE Debug )
#set(DEPENDENCIES realsense2 )
target_link_libraries(measure_distance librealsense2.so)- 编译测距程序，生成可执行文件
mkdir build && cd build && cmake .. && make -j88.3 error
terminate called after throwing an instance of 'rs2::invalid_value_error'
what(): API version mismatch: librealsense.so was compiled with API version 2.21.0 but the application was compiled with 2.19.1! Make sure correct version of the library is installed (make install)
Linux系统下的编译，编写CMakeLists.txt时，需要找到程序需要的库文件、头文件，但库/头文件需要对应到同一版本。如果在CMakeLists.txt文件中使用了link_directories，则编译器会去系统默认的头文件位置(其中一个为/usr/local/include/)查找需要的头文件。这个错误就是，我手动编译出来的API是2.21.0，但是未包含对应版本的头文件，编译过程自动到/usr/local/include/目录下查找对应API的头文件，导致两者版本不匹配。
9.编译ffmpeg
参考网站：http://www.cnblogs.com/CoderTian/p/6655568.html
下载：
git clone https://github.com/FFmpeg/FFmpeg.git配置命令：(编译至host目录下)
./configure --prefix=host --enable-shared --disable-static --disable-doc 编译：make -j8 && make install
host目录下编译产生:
├── bin
│ ├── ffmpeg
│ └── ffprobe
├── include
│ ├── libavcodec
│ │ ├── ac3_parser.h
│ │ ├── adts_parser.h
│ │ ├── avcodec.h
……
│ ├── libavdevice
│ │ ├── avdevice.h
│ │ └── version.h
│ ├── libavfilter
│ │ ├── avfilter.h
│ │ ├── buffersink.h
│ │ ├── buffersrc.h
│ │ └── version.h
│ ├── libavformat
│ │ ├── avformat.h
│ │ ├── avio.h
│ │ └── version.h
│ ├── libavutil
│ │ ├── adler32.h
│ │ ├── aes_ctr.h
│ │ ├── aes.h
……
│ ├── libswresample
│ │ ├── swresample.h
│ │ └── version.h
│ └── libswscale
│ ├── swscale.h
│ └── version.h
├── lib
│ ├── libavcodec.so -> libavcodec.so.58.52.101
│ ├── libavcodec.so.58 -> libavcodec.so.58.52.101
│ ├── libavcodec.so.58.52.101
│ ├── libavdevice.so -> libavdevice.so.58.7.100
│ ├── libavdevice.so.58 -> libavdevice.so.58.7.100
│ ├── libavdevice.so.58.7.100
│ ├── libavfilter.so -> libavfilter.so.7.53.100
│ ├── libavfilter.so.7 -> libavfilter.so.7.53.100
……
编译muxing.cpp
从doc/examples复制到host目录下，后缀改为cpp，并将头文件改为：
#define __STDC_CONSTANT_MACROS
#include
#ifdef __cplusplus
extern "C"
{
#endif
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#include
#ifdef __cplusplus
};
#endif host目录下建立CMakeLists.txt
project(ffmpeg_muxing)
cmake_minimum_required(VERSION 2.8)
cmake_policy(SET CMP0015 NEW)
include_directories(include)
set(CMAKE_CXX_FLAGS "-std=c++11")
##添加头文件
link_directories(lib) #添加复制出来的链接库
add_executable(push_flow_muxing muxing.cpp)
#添加后可进行调试
set( CMAKE_BUILD_TYPE Debug )
target_link_libraries(push_flow_muxing libavcodec.so libavformat.so libavutil.so libswresample.so libswscale.so)编译muxing.cpp的可执行文件：
mkdir build && cd build && cmake .. && make 修改bug参考网站 ：https://github.com/usc-imi/aeo-light/issues/8
host/include/libavutil/error.h:121:95: error: taking address of temporary array av_make_error_string((char[AV_ERROR_MAX_STRING_SIZE]){0}, AV_ERROR_MAX_STRING_SIZE, errnum)host/include/libavutil/error.h:121:95: error: taking address of temporary array av_make_error_string((char[AV_ERROR_MAX_STRING_SIZE]){0}, AV_ERROR_MAX_STRING_SIZE, errnum)如果需要将自己写的代码编译成.so文件并链接到可执行文件，只需要将CMakeLists.txt改为以下方式：(编译完成之后会生成一个liblive_libs.so文件以及一个可执行文件app)
………………
set(src_files
src/1.cpp
src/2.cpp
src/3.cpp
src/4.cpp
)
add_library(live_libs SHARED ${src_files})
aux_source_directory(./src DIR_SRCS)
add_executable(app ${DIR_SRCS})
target_link_libraries(app ${live_libs})10.Rock Pi 4B(RK3399)配置Paddle-Lite
参考网站1：https://paddlepaddle.github.io/Paddle-Lite/
参考网站2：https://github.com/YunYang1994/tensorflow-yolov3
step1 配置基本工具：
- gcc、g++、git、make、wget、python
- cmake（建议使用3.10或以上版本）
step2 将tensorflow-yolov3模型转为Paddle-Lite兼容的模型
转换工具：X2Paddle
模型来源：模型格式为(.pb)
- 按照参考网站2配置完成tensorflow框架下的yolov3，part 1. Quick start。（注意：下载的模型放在checkpoint路径下；运行过程可能存在一些问题，core/utils.py中的scale = min(iw/w, ih/h)需要改为scale = min(float(iw)/w, float(ih)/h)）
- 按照X2Paddle转换(.pb)为PaddlePaddle深度学习框架产出的模型格式：x2paddle --framework=tensorflow --model=tf_model.pb --save_dir=pd_model（可能需要输入tensor的shape，以及模型转换失败，则参考FAQ）
- 模型优化：我是直接在x86_64普通电脑下转的，因此下载工具链接：./model_optimize_tool.x86_64-linux-gnu --model_dir=inference_model/ –--optimize_out_type=naive_buffer --optimize_out =optimize_model_path/（在RK3399上armv8系统架构下载链接。）
step3 编译Paddle-Lite-Demo：（step2其实也不是必须的。step3可能需要编译opencv交叉编译，手动编译）
$ cd Paddle-Lite-Demo/PaddleLite-armlinux-demo/image_classification_demo
$ ./run.sh armv8 # RK3399
实现的演示效果如下：
| 分类cat | 目标检测 |
11.Rock Pi 4B(RK3399)配置TensorFlow-Lite v1.14
参考网站1：https://www.wandouip.com/t5i214530/
参考网站2：https://blog.csdn.net/computerme/article/details/80345065
参考网站3：https://github.com/tensorflow/tensorflow/issues/32073
参考网站4：https://tensorflow.google.cn/lite/guide/build_arm64
硬件内核(uname -a)：Linux linaro-alip 4.4.154-90-rockchip-ga14f6502e045 #22 SMP Tue Jul 30 10:32:28 UTC 2019 aarch64 GNU/Linux
交叉编译（在x86_64主机上编译，执行直接拷贝至rockpi）：
11.1 clone或者下载TensorFlow代码(最好选定为r1.14分支)
git clone [email protected]:tensorflow/tensorflow.git11.2 根据参考网站4配置交叉编译环境
sudo apt-get install crossbuild-essential-arm6411.3 下载TensorFlow所需依赖（TensorFlow根目录下运行，国内服务器可能会断开连接，则参考网站1，下载后解压至tensorflow/lite/tools/make/downloads目录下）
./tensorflow/lite/tools/make/download_dependencies.sh11.4 编译（TensorFlow根目录下运行）
./tensorflow/lite/tools/make/build_aarch64_lib.sh以上步骤能成功编译出一个静态库：tensorflow/lite/tools/make/gen/aarch64_armv8-a/lib/libtensorflow-lite.a
11.5 编译label_image分类Demo
参考网站1、网站3修改tensorflow/lite/tools/make/Makefile文件，修改后的文件如下：
①有MINIMAL的地方，都仿照添加LABEL_IMAGE信息
②添加tensorflow/lite/tools/evaluation/utils.cc文件（否则会报错undefined reference："tflite::evaluation::CreateGPUDelegate(tflite::FlateBufferModel)"）
# Make uses /bin/sh by default, which is incompatible with the bashisms seen
# below.
SHELL := /bin/bash
# Find where we're running from, so we can store generated files here.
ifeq ($(origin MAKEFILE_DIR), undefined)
MAKEFILE_DIR := $(shell dirname $(realpath $(lastword $(MAKEFILE_LIST))))
endif
# Try to figure out the host system
HOST_OS :=
ifeq ($(OS),Windows_NT)
HOST_OS = windows
else
UNAME_S := $(shell uname -s)
ifeq ($(UNAME_S),Linux)
HOST_OS := linux
endif
ifeq ($(UNAME_S),Darwin)
HOST_OS := osx
endif
endif
HOST_ARCH := $(shell if uname -m | grep -q i[345678]86; then echo x86_32; else uname -m; fi)
# Override these on the make command line to target a specific architecture. For example:
# make -f tensorflow/lite/tools/make/Makefile TARGET=rpi TARGET_ARCH=armv7l
TARGET := $(HOST_OS)
TARGET_ARCH := $(HOST_ARCH)
INCLUDES := \
-I. \
-I$(MAKEFILE_DIR)/../../../../../ \
-I$(MAKEFILE_DIR)/../../../../../../ \
-I$(MAKEFILE_DIR)/downloads/ \
-I$(MAKEFILE_DIR)/downloads/eigen \
-I$(MAKEFILE_DIR)/downloads/absl \
-I$(MAKEFILE_DIR)/downloads/gemmlowp \
-I$(MAKEFILE_DIR)/downloads/neon_2_sse \
-I$(MAKEFILE_DIR)/downloads/farmhash/src \
-I$(MAKEFILE_DIR)/downloads/flatbuffers/include \
-I$(OBJDIR)
# This is at the end so any globally-installed frameworks like protobuf don't
# override local versions in the source tree.
INCLUDES += -I/usr/local/include
# These are the default libraries needed, but they can be added to or
# overridden by the platform-specific settings in target makefiles.
LIBS := \
-lstdc++ \
-lpthread \
-lm \
-lz
# There are no rules for compiling objects for the host system (since we don't
# generate things like the protobuf compiler that require that), so all of
# these settings are for the target compiler.
CXXFLAGS := -O3 -DNDEBUG -fPIC
CXXFLAGS += $(EXTRA_CXXFLAGS)
CFLAGS := ${CXXFLAGS}
CXXFLAGS += --std=c++11
LDOPTS := -L/usr/local/lib
ARFLAGS := -r
TARGET_TOOLCHAIN_PREFIX :=
CC_PREFIX :=
ifeq ($(HOST_OS),windows)
CXXFLAGS += -fext-numeric-literals -D__LITTLE_ENDIAN__
endif
# This library is the main target for this makefile. It will contain a minimal
# runtime that can be linked in to other programs.
LIB_NAME := libtensorflow-lite.a
# Benchmark static library and binary
BENCHMARK_LIB_NAME := benchmark-lib.a
BENCHMARK_BINARY_NAME := benchmark_model
BENCHMARK_PERF_OPTIONS_BINARY_NAME := benchmark_model_performance_options
# A small example program that shows how to link against the library.
MINIMAL_SRCS := \
tensorflow/lite/examples/minimal/minimal.cc
LABEL_IMAGE_SRCS := \
tensorflow/lite/examples/label_image/label_image.cc\
tensorflow/lite/examples/label_image/bitmap_helpers.cc\
tensorflow/lite/tools/evaluation/utils.cc
# What sources we want to compile, must be kept in sync with the main Bazel
# build files.
PROFILER_SRCS := \
tensorflow/lite/profiling/memory_info.cc \
tensorflow/lite/profiling/time.cc
PROFILE_SUMMARIZER_SRCS := \
tensorflow/lite/profiling/profile_summarizer.cc \
tensorflow/core/util/stats_calculator.cc
CMD_LINE_TOOLS_SRCS := \
tensorflow/lite/tools/command_line_flags.cc
CORE_CC_ALL_SRCS := \
$(wildcard tensorflow/lite/*.cc) \
$(wildcard tensorflow/lite/*.c) \
$(wildcard tensorflow/lite/c/*.c) \
$(wildcard tensorflow/lite/core/*.cc) \
$(wildcard tensorflow/lite/core/api/*.cc) \
$(wildcard tensorflow/lite/experimental/resource_variable/*.cc) \
$(wildcard tensorflow/lite/experimental/ruy/*.cc)
ifneq ($(BUILD_TYPE),micro)
CORE_CC_ALL_SRCS += \
$(wildcard tensorflow/lite/kernels/*.cc) \
$(wildcard tensorflow/lite/kernels/internal/*.cc) \
$(wildcard tensorflow/lite/kernels/internal/optimized/*.cc) \
$(wildcard tensorflow/lite/kernels/internal/reference/*.cc) \
$(PROFILER_SRCS) \
tensorflow/lite/tools/make/downloads/farmhash/src/farmhash.cc \
tensorflow/lite/tools/make/downloads/fft2d/fftsg.c \
tensorflow/lite/tools/make/downloads/flatbuffers/src/util.cpp
CORE_CC_ALL_SRCS += \
$(shell find tensorflow/lite/tools/make/downloads/absl/absl/ \
-type f -name \*.cc | grep -v test | grep -v benchmark | grep -v synchronization | grep -v debugging)
endif
# Remove any duplicates.
CORE_CC_ALL_SRCS := $(sort $(CORE_CC_ALL_SRCS))
CORE_CC_EXCLUDE_SRCS := \
$(wildcard tensorflow/lite/*test.cc) \
$(wildcard tensorflow/lite/*/*test.cc) \
$(wildcard tensorflow/lite/*/*/benchmark.cc) \
$(wildcard tensorflow/lite/*/*/example*.cc) \
$(wildcard tensorflow/lite/*/*/test*.cc) \
$(wildcard tensorflow/lite/*/*/*test.cc) \
$(wildcard tensorflow/lite/*/*/*/*test.cc) \
$(wildcard tensorflow/lite/kernels/*test_main.cc) \
$(wildcard tensorflow/lite/kernels/*test_util*.cc) \
$(MINIMAL_SRCS)\
$(LABEL_IMAGE_SRCS)
BUILD_WITH_MMAP ?= true
ifeq ($(BUILD_TYPE),micro)
BUILD_WITH_MMAP=false
endif
ifeq ($(BUILD_TYPE),windows)
BUILD_WITH_MMAP=false
endif
ifeq ($(BUILD_WITH_MMAP),true)
CORE_CC_EXCLUDE_SRCS += tensorflow/lite/mmap_allocation.cc
else
CORE_CC_EXCLUDE_SRCS += tensorflow/lite/mmap_allocation_disabled.cc
endif
BUILD_WITH_NNAPI ?= true
ifeq ($(BUILD_TYPE),micro)
BUILD_WITH_NNAPI=false
endif
ifeq ($(TARGET),windows)
BUILD_WITH_NNAPI=false
endif
ifeq ($(TARGET),ios)
BUILD_WITH_NNAPI=false
endif
ifeq ($(TARGET),rpi)
BUILD_WITH_NNAPI=false
endif
ifeq ($(TARGET),generic-aarch64)
BUILD_WITH_NNAPI=false
endif
ifeq ($(BUILD_WITH_NNAPI),true)
CORE_CC_ALL_SRCS += tensorflow/lite/delegates/nnapi/nnapi_delegate.cc
CORE_CC_ALL_SRCS += tensorflow/lite/delegates/nnapi/quant_lstm_sup.cc
CORE_CC_ALL_SRCS += tensorflow/lite/nnapi/nnapi_implementation.cc
CORE_CC_ALL_SRCS += tensorflow/lite/nnapi/nnapi_util.cc
LIBS += -lrt
else
CORE_CC_ALL_SRCS += tensorflow/lite/delegates/nnapi/nnapi_delegate_disabled.cc
CORE_CC_ALL_SRCS += tensorflow/lite/nnapi/nnapi_implementation_disabled.cc
endif
ifeq ($(TARGET),ios)
CORE_CC_EXCLUDE_SRCS += tensorflow/lite/minimal_logging_android.cc
CORE_CC_EXCLUDE_SRCS += tensorflow/lite/minimal_logging_default.cc
else
CORE_CC_EXCLUDE_SRCS += tensorflow/lite/minimal_logging_android.cc
CORE_CC_EXCLUDE_SRCS += tensorflow/lite/minimal_logging_ios.cc
endif
# Filter out all the excluded files.
TF_LITE_CC_SRCS := $(filter-out $(CORE_CC_EXCLUDE_SRCS), $(CORE_CC_ALL_SRCS))
# Benchmark sources
BENCHMARK_SRCS_DIR := tensorflow/lite/tools/benchmark
EVALUATION_UTILS_SRCS := \
tensorflow/lite/tools/evaluation/utils.cc
BENCHMARK_ALL_SRCS := \
$(wildcard $(BENCHMARK_SRCS_DIR)/*.cc) \
$(PROFILE_SUMMARIZER_SRCS) \
$(CMD_LINE_TOOLS_SRCS) \
$(EVALUATION_UTILS_SRCS)
BENCHMARK_MAIN_SRC := $(BENCHMARK_SRCS_DIR)/benchmark_main.cc
BENCHMARK_PERF_OPTIONS_SRC := \
$(BENCHMARK_SRCS_DIR)/benchmark_tflite_performance_options_main.cc
BENCHMARK_LIB_SRCS := $(filter-out \
$(wildcard $(BENCHMARK_SRCS_DIR)/*_test.cc) \
$(BENCHMARK_MAIN_SRC) \
$(BENCHMARK_PERF_OPTIONS_SRC) \
$(BENCHMARK_SRCS_DIR)/benchmark_plus_flex_main.cc, \
$(BENCHMARK_ALL_SRCS))
# These target-specific makefiles should modify or replace options like
# CXXFLAGS or LIBS to work for a specific targetted architecture. All logic
# based on platforms or architectures should happen within these files, to
# keep this main makefile focused on the sources and dependencies.
include $(wildcard $(MAKEFILE_DIR)/targets/*_makefile.inc)
ALL_SRCS := \
$(MINIMAL_SRCS) \
$(LABEL_IMAGE_SRCS) \
$(PROFILER_SRCS) \
$(PROFILER_SUMMARIZER_SRCS) \
$(TF_LITE_CC_SRCS) \
$(BENCHMARK_LIB_SRCS) \
$(CMD_LINE_TOOLS_SRCS)
# Where compiled objects are stored.
GENDIR := $(MAKEFILE_DIR)/gen/$(TARGET)_$(TARGET_ARCH)/
OBJDIR := $(GENDIR)obj/
BINDIR := $(GENDIR)bin/
LIBDIR := $(GENDIR)lib/
LIB_PATH := $(LIBDIR)$(LIB_NAME)
BENCHMARK_LIB := $(LIBDIR)$(BENCHMARK_LIB_NAME)
BENCHMARK_BINARY := $(BINDIR)$(BENCHMARK_BINARY_NAME)
BENCHMARK_PERF_OPTIONS_BINARY := $(BINDIR)$(BENCHMARK_PERF_OPTIONS_BINARY_NAME)
MINIMAL_BINARY := $(BINDIR)minimal
LABEL_IMAGE_BINARY := $(BINDIR)label_image
CXX := $(CC_PREFIX)${TARGET_TOOLCHAIN_PREFIX}g++
CC := $(CC_PREFIX)${TARGET_TOOLCHAIN_PREFIX}gcc
AR := $(CC_PREFIX)${TARGET_TOOLCHAIN_PREFIX}ar
MINIMAL_OBJS := $(addprefix $(OBJDIR), \
$(patsubst %.cc,%.o,$(patsubst %.c,%.o,$(MINIMAL_SRCS))))
LABEL_IMAGE_OBJS := $(addprefix $(OBJDIR), \
$(patsubst %.cc,%.o,$(patsubst %.c,%.o,$(LABEL_IMAGE_SRCS))))
LIB_OBJS := $(addprefix $(OBJDIR), \
$(patsubst %.cc,%.o,$(patsubst %.c,%.o,$(patsubst %.cpp,%.o,$(TF_LITE_CC_SRCS)))))
BENCHMARK_MAIN_OBJ := $(addprefix $(OBJDIR), \
$(patsubst %.cc,%.o,$(patsubst %.c,%.o,$(BENCHMARK_MAIN_SRC))))
BENCHMARK_PERF_OPTIONS_OBJ := $(addprefix $(OBJDIR), \
$(patsubst %.cc,%.o,$(patsubst %.c,%.o,$(BENCHMARK_PERF_OPTIONS_SRC))))
BENCHMARK_LIB_OBJS := $(addprefix $(OBJDIR), \
$(patsubst %.cc,%.o,$(patsubst %.c,%.o,$(BENCHMARK_LIB_SRCS))))
# For normal manually-created TensorFlow Lite C++ source files.
$(OBJDIR)%.o: %.cc
@mkdir -p $(dir $@)
$(CXX) $(CXXFLAGS) $(INCLUDES) -c $< -o $@
# For normal manually-created TensorFlow Lite C source files.
$(OBJDIR)%.o: %.c
@mkdir -p $(dir $@)
$(CC) $(CFLAGS) $(INCLUDES) -c $< -o $@
$(OBJDIR)%.o: %.cpp
@mkdir -p $(dir $@)
$(CXX) $(CXXFLAGS) $(INCLUDES) -c $< -o $@
# The target that's compiled if there's no command-line arguments.
all: $(LIB_PATH) $(MINIMAL_BINARY) $(BENCHMARK_BINARY) $(BENCHMARK_PERF_OPTIONS_BINARY) $(LABEL_IMAGE_BINARY)
# The target that's compiled for micro-controllers
micro: $(LIB_PATH)
# Hack for generating schema file bypassing flatbuffer parsing
tensorflow/lite/schema/schema_generated.h:
@cp -u tensorflow/lite/schema/schema_generated.h.OPENSOURCE tensorflow/lite/schema/schema_generated.h
# Gathers together all the objects we've compiled into a single '.a' archive.
$(LIB_PATH): tensorflow/lite/schema/schema_generated.h $(LIB_OBJS)
@mkdir -p $(dir $@)
$(AR) $(ARFLAGS) $(LIB_PATH) $(LIB_OBJS)
lib: $(LIB_PATH)
$(MINIMAL_BINARY): $(MINIMAL_OBJS) $(LIB_PATH)
@mkdir -p $(dir $@)
$(CXX) $(CXXFLAGS) $(INCLUDES) \
-o $(MINIMAL_BINARY) $(MINIMAL_OBJS) \
$(LIBFLAGS) $(LIB_PATH) $(LDFLAGS) $(LIBS)
$(LABEL_IMAGE_BINARY): $(LABEL_IMAGE_OBJS) $(LIB_PATH)
@mkdir -p $(dir $@)
$(CXX) $(CXXFLAGS) $(INCLUDES) \
-o $(LABEL_IMAGE_BINARY) $(LABEL_IMAGE_OBJS) \
$(LIBFLAGS) $(LIB_PATH) $(LDFLAGS) $(LIBS)
minimal: $(MINIMAL_BINARY)
label_image: $(LABEL_IMAGE_BINARY)
$(BENCHMARK_LIB) : $(LIB_PATH) $(BENCHMARK_LIB_OBJS)
@mkdir -p $(dir $@)
$(AR) $(ARFLAGS) $(BENCHMARK_LIB) $(LIB_OBJS) $(BENCHMARK_LIB_OBJS)
benchmark_lib: $(BENCHMARK_LIB)
$(BENCHMARK_BINARY) : $(BENCHMARK_MAIN_OBJ) $(BENCHMARK_LIB)
@mkdir -p $(dir $@)
$(CXX) $(CXXFLAGS) $(INCLUDES) \
-o $(BENCHMARK_BINARY) $(BENCHMARK_MAIN_OBJ) \
$(LIBFLAGS) $(BENCHMARK_LIB) $(LDFLAGS) $(LIBS)
$(BENCHMARK_PERF_OPTIONS_BINARY) : $(BENCHMARK_PERF_OPTIONS_OBJ) $(BENCHMARK_LIB)
@mkdir -p $(dir $@)
$(CXX) $(CXXFLAGS) $(INCLUDES) \
-o $(BENCHMARK_PERF_OPTIONS_BINARY) $(BENCHMARK_PERF_OPTIONS_OBJ) \
$(LIBFLAGS) $(BENCHMARK_LIB) $(LDFLAGS) $(LIBS)
benchmark: $(BENCHMARK_BINARY) $(BENCHMARK_PERF_OPTIONS_BINARY)
libdir:
@echo $(LIBDIR)
# Gets rid of all generated files.
clean:
rm -rf $(MAKEFILE_DIR)/gen
# Gets rid of target files only, leaving the host alone. Also leaves the lib
# directory untouched deliberately, so we can persist multiple architectures
# across builds for iOS and Android.
cleantarget:
rm -rf $(OBJDIR)
rm -rf $(BINDIR)
$(DEPDIR)/%.d: ;
.PRECIOUS: $(DEPDIR)/%.d
-include $(patsubst %,$(DEPDIR)/%.d,$(basename $(ALL_SRCS)))
按照11.4再次编译，最终编译生成：tensorflow/lite/tools/make/gen/aarch64_armv8-a/bin/label_image
TensorFlow根目录下使用以下命令可查看这个可执行文件的运行内核环境
file tensorflow/lite/tools/make/gen/aarch64_armv8-a/bin/label_image输出：ELF 64-bit LSB executable, ARM aarch64, version 1 (GNU/Linux), dynamically linked, interpreter /lib/ld-, for GNU/Linux 3.7.0, BuildID[sha1]=41093a674cedba6b014bd47da95a0a09b9aa9aa8, not stripped
11.6 在网站中下载model和labels文件，并拷贝模型和tensorflow/lite/tools/make/gen/aarch64_armv8-a/下的所有文件至rockpi中，运行命令：
./label_image -v 1 -m ./mobilenet_v1_1.0_224_quant.tflite -i ./grace_hopper.bmp -l ./labels_mobilenet_quant_v1_224.txt 输出：average time: 113.777 ms
0.780392: 653 military uniform
0.105882: 907 Windsor tie
0.0156863: 458 bow tie
0.0117647: 466 bulletproof vest
0.00784314: 835 suit
11.7 测试目标检测模型
参考网站：https://github.com/tensorflow/examples/tree/master/lite/examples/object_detection/raspberry_pi
①rockpi的内核是aarch64，因此下载网站中的tflite_runtime-1.14.0-cp35-cp35m-linux_aarch64.whl，并拷贝至rockpi中，使用pip install tflite_runtime-1.14.0-cp35-cp35m-linux_aarch64.whl安装；
②按照网站中的配置过程，下载代码，若不配置摄像头，仅测试单张图片，则修改网站中的detect_picamera.py
# python3
#
# Copyright 2019 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# https://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""Example using TF Lite to detect objects with the Raspberry Pi camera."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import argparse
import io
import re
import time
from annotation import Annotator
import numpy as np
from PIL import Image,ImageDraw
from tflite_runtime.interpreter import Interpreter
CAMERA_WIDTH = 640
CAMERA_HEIGHT = 480
def load_labels(path):
"""Loads the labels file. Supports files with or without index numbers."""
with open(path, 'r', encoding='utf-8') as f:
lines = f.readlines()
labels = {}
for row_number, content in enumerate(lines):
pair = re.split(r'[:\s]+', content.strip(), maxsplit=1)
if len(pair) == 2 and pair[0].strip().isdigit():
labels[int(pair[0])] = pair[1].strip()
else:
labels[row_number] = pair[0].strip()
return labels
def set_input_tensor(interpreter, image):
"""Sets the input tensor."""
tensor_index = interpreter.get_input_details()[0]['index']
input_tensor = interpreter.tensor(tensor_index)()[0]
input_tensor[:, :] = image
def get_output_tensor(interpreter, index):
"""Returns the output tensor at the given index."""
output_details = interpreter.get_output_details()[index]
tensor = np.squeeze(interpreter.get_tensor(output_details['index']))
return tensor
def detect_objects(interpreter, image, threshold):
"""Returns a list of detection results, each a dictionary of object info."""
set_input_tensor(interpreter, image)
interpreter.invoke()
#import pdb;pdb.set_trace()
# Get all output details
boxes = get_output_tensor(interpreter, 0)
classes = get_output_tensor(interpreter, 1)
scores = get_output_tensor(interpreter, 2)
count = int(get_output_tensor(interpreter, 3))
results = []
for i in range(count):
if scores[i] >= threshold:
result = {
'bounding_box': boxes[i],
'class_id': classes[i],
'score': scores[i]
}
results.append(result)
return results
def annotate_objects(image, results, labels):
"""Draws the bounding box and label for each object in the results."""
image = image.resize(
(CAMERA_WIDTH, CAMERA_HEIGHT), Image.ANTIALIAS)
draw = ImageDraw.Draw(image)
for obj in results:
# Convert the bounding box figures from relative coordinates
# to absolute coordinates based on the original resolution
ymin, xmin, ymax, xmax = obj['bounding_box']
xmin = int(xmin * CAMERA_WIDTH)
xmax = int(xmax * CAMERA_WIDTH)
ymin = int(ymin * CAMERA_HEIGHT)
ymax = int(ymax * CAMERA_HEIGHT)
# Overlay the box, label, and score on the camera preview
draw.rectangle((xmin,ymin,xmax,ymax),None,'red')
draw.text([xmin, ymin],
'%s\n%.2f' % (labels[obj['class_id']], obj['score']))
image.save('car_out.jpg','JPEG')
def main():
parser = argparse.ArgumentParser(
formatter_class=argparse.ArgumentDefaultsHelpFormatter)
parser.add_argument(
'--model', help='File path of .tflite file.', required=True)
parser.add_argument(
'--labels', help='File path of labels file.', required=True)
parser.add_argument(
'--threshold',
help='Score threshold for detected objects.',
required=False,
type=float,
default=0.1)
args = parser.parse_args()
labels = load_labels(args.labels)
interpreter = Interpreter(args.model)
interpreter.allocate_tensors()
_, input_height, input_width, _ = interpreter.get_input_details()[0]['shape']
img_d = Image.open('car.jpg').convert('RGB')
image = img_d.resize(
(input_width, input_height), Image.ANTIALIAS)
start_time = time.monotonic()
results = detect_objects(interpreter, image, args.threshold)
elapsed_ms = (time.monotonic() - start_time) * 1000
annotate_objects(img_d, results, labels)
if __name__ == '__main__':
main()
执行下面的命令即可
python3 detect_picamera.py --model detect.tflite --labels coco_labels.txt