paddlepaddle/flycv

FlyCV is a high-performance library for processing computer visual tasks.

flycv's Issues

flycv 在rv1126 处理图片也不是很快 感觉跟 opecv 处理时间一样

有什么办法证明 flycv 在运行吗
现在 再用 fastdeploy

[INFO] fastdeploy/vision/common/processors/base.cc(154)::EnableFlyCV Will change to use image processing library ProcLib::FLYCV

There are 2 problems with the "extract_channel" interface

1. first
   opencv设计该函数的时候没有用引用: void cv::extractChannel(InputArray _src, OutputArray _dst, int coi)
   flyCV的定义是 FCV_API int extract_channel(Mat& src,Mat& dst, int index);
   问题，下面写法会报错：error: undefined reference to 'fcv::extract_channel(fcv::Mat&, fcv::Mat, int)'
   g_fcv_ns::extract_channel(preimg, g_fcv_ns::Mat(cols, rows,g_fcv_ns::FCVImageType::GRAY_F32,data +i * rows * cols), i);

2. Second
   目前只支持u8类型，导致只支持int8模型，flyCV的巨大威力被卡脖子了。

windows编译必须先编译静态库才能编译动态库

如果先编译动态库：

mkdir build
cd build
cmake .. -GNinja -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=./Release -DBUILD_SHARED_LIBS=ON -DBUILD_TEST=OFF -DWITH_LIB_PNG=ON -DWITH_LIB_JPEG_TURBO=ON -DBUILD_C=ON
ninja all

则会报如下错：

C:\_\code\FlyCV\build>ninja all
ninja: error: 'third_party/libjpeg-turbo/output/lib/turbojpeg-static.lib', needed by 'modules/flycv_shared.dll', missing and no known rule to make it

必须先编译静态库：

C:\_\code\FlyCV\build>cmake .. -GNinja -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=./Release -DBUILD_SHARED_LIBS=OFF -DBUILD_TEST=OFF -DWITH_LIB_PNG=ON -DWITH_LIB_JPEG_TURBO=ON -DBUILD_C=ON
ninja all

静态库编译成功之后，才能正常编译动态库。

希望新增video io

flycv与opencv同时读图，opencv读图失败，数据为空

测试环境：台式ubuntu20.04，opencv3.4.14，flycv1.0.0
效果：fcv和opencv各自注释掉对方，独立的运行没问题，但是一起执行，opencv读图失败，图像数据为空

测试参考代码：
#include
#include <opencv2/opencv.hpp>
#include <dirent.h>
#include "flycv.h"
#include "pthread.h"
void calc_param(float& ratio, float& dww, float& dhh, int org_width, int org_height, int new_width, int new_hieght,
bool is_auto, bool scale_fill, bool scale_up, int stride) {
float r = std::min(new_width * 1.0 / org_width, new_hieght * 1.0 / org_height);
if (!scale_up)
r = std::min(r, float(1.0));

// ratio_w_ = r;
// ratio_h_ = r;
ratio = r;
int new_unpad_w_ = int(org_width * r);
int new_unpad_h_ = int(org_height * r);
int dw = new_width - new_unpad_w_;
int dh = new_hieght - new_unpad_h_;
if (is_auto) {
dw = dw % stride;
dh = dh % stride;
} else if (scale_fill) {
dw = 0, dh = 0;
new_unpad_w_ = new_width;
new_unpad_h_ = new_hieght;
float ratio_w_ = new_width * 1.0 / org_width;
float ratio_h_ = new_hieght * 1.0 / org_height;
}
// dw_ = dw / 2.0;
// dh_ = dh / 2.0;
dww = dw / 2.0;
dhh = dh / 2.0;
}
int main(int argc, char** argv) {

// // fcv的处理
fcv::Mat src_fcv = fcv::imread("../100.jpg");
std::cout << src_fcv.width() << " " << src_fcv.height() << std::endl;
float ratio_fcv, dww_fcv, dhh_fcv;
calc_param(ratio_fcv, dww_fcv, dhh_fcv, src_fcv.width(), src_fcv.height(), 640, 640, false, false, true, 2);
fcv::Mat resize_im_fcv;
double r_fcv = ratio_fcv;
fcv::resize(src_fcv, resize_im_fcv, fcv::Size(), r_fcv, r_fcv);
fcv::Mat padding_img_fcv;
int top_fcv = int(round(dhh_fcv - 0.1));
int bottom_fcv = int(round(dhh_fcv + 0.1));
int left_fcv = int(round(dww_fcv - 0.1));
int right_fcv = int(round(dww_fcv + 0.1));
fcv::Scalar color_fcv(114, 114, 114);
fcv::copy_make_border(resize_im_fcv, padding_img_fcv, top_fcv, bottom_fcv, left_fcv, right_fcv, fcv::BorderType::BORDER_CONSTANT, color_fcv);
fcv::imwrite("padding_img_fcv.jpg", padding_img_fcv);

// //opencv的处理
cv::Mat src_cv = cv::imread("../200.jpg");
std::cout << src_cv.cols << "  " << src_cv.rows << std::endl;
float ratio, dww, dhh;
calc_param(ratio, dww, dhh, src_cv.cols, src_cv.rows, 640, 640, false, false, true, 2);
cv::Mat resize_im_cv;
cv::resize(src_cv, resize_im_cv, cv::Size(), ratio, ratio);
cv::Mat padding_img_cv;
int top = int(round(dhh - 0.1));
int bottom = int(round(dhh + 0.1));
int left = int(round(dww - 0.1));
int right = int(round(dww + 0.1));
cv::Scalar color_cv(114, 114, 114);
cv::copyMakeBorder(resize_im_cv, padding_img_cv, top, bottom, left, right, cv::BORDER_CONSTANT, color_cv);
cv::imwrite("padding_img_cv.jpg", padding_img_cv);
return 0;

}

官方为什么不提供1.2.0版本的windowsx64和x86的flycv库

我看了下1.2.0版本所有平台都有唯独预编译windows没有

-DWITH_LIB_PNG在Windows下无法编译

环境：VS2022
命令行：

cmake .. -GNinja -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=./Release -DBUILD_SHARED_LIBS=ON -DBUILD_TEST=OFF -DWITH_LIB_PNG=ON -DWITH_LIB_JPEG_TURBO=ON
ninja all

报错信息：

C:\_\3rd\FlyCV\build>ninja all
[1/88] Performing install step for 'zlib'
[0/1] Install the project...
-- Install configuration: "Release"
-- Installing: C:/_/3rd/FlyCV/build/third_party/zlib/output/lib/zlib.lib
-- Installing: C:/_/3rd/FlyCV/build/third_party/zlib/output/bin/zlib.dll
-- Installing: C:/_/3rd/FlyCV/build/third_party/zlib/output/lib/zlibstatic.lib
-- Installing: C:/_/3rd/FlyCV/build/third_party/zlib/output/include/zconf.h
-- Installing: C:/_/3rd/FlyCV/build/third_party/zlib/output/include/zlib.h
-- Installing: C:/_/3rd/FlyCV/build/third_party/zlib/output/share/man/man3/zlib.3
-- Installing: C:/_/3rd/FlyCV/build/third_party/zlib/output/share/pkgconfig/zlib.pc
[2/88] Performing install step for 'libjpeg-turbo'
[0/1] Install the project...
-- Install configuration: "Release"
-- Installing: C:/_/3rd/FlyCV/build/third_party/libjpeg-turbo/output/lib/turbojpeg-static.lib
-- Installing: C:/_/3rd/FlyCV/build/third_party/libjpeg-turbo/output/bin/tjbench.exe
-- Installing: C:/_/3rd/FlyCV/build/third_party/libjpeg-turbo/output/include/turbojpeg.h
-- Installing: C:/_/3rd/FlyCV/build/third_party/libjpeg-turbo/output/lib/jpeg-static.lib
-- Installing: C:/_/3rd/FlyCV/build/third_party/libjpeg-turbo/output/bin/cjpeg.exe
-- Installing: C:/_/3rd/FlyCV/build/third_party/libjpeg-turbo/output/bin/djpeg.exe
-- Installing: C:/_/3rd/FlyCV/build/third_party/libjpeg-turbo/output/bin/jpegtran.exe
-- Installing: C:/_/3rd/FlyCV/build/third_party/libjpeg-turbo/output/bin/rdjpgcom.exe
-- Installing: C:/_/3rd/FlyCV/build/third_party/libjpeg-turbo/output/bin/wrjpgcom.exe
-- Installing: C:/_/3rd/FlyCV/build/third_party/libjpeg-turbo/output/share/doc/libjpeg-turbo/README.ijg
-- Installing: C:/_/3rd/FlyCV/build/third_party/libjpeg-turbo/output/share/doc/libjpeg-turbo/README.md
-- Installing: C:/_/3rd/FlyCV/build/third_party/libjpeg-turbo/output/share/doc/libjpeg-turbo/example.txt
-- Installing: C:/_/3rd/FlyCV/build/third_party/libjpeg-turbo/output/share/doc/libjpeg-turbo/tjexample.c
-- Installing: C:/_/3rd/FlyCV/build/third_party/libjpeg-turbo/output/share/doc/libjpeg-turbo/libjpeg.txt
-- Installing: C:/_/3rd/FlyCV/build/third_party/libjpeg-turbo/output/share/doc/libjpeg-turbo/structure.txt
-- Installing: C:/_/3rd/FlyCV/build/third_party/libjpeg-turbo/output/share/doc/libjpeg-turbo/usage.txt
-- Installing: C:/_/3rd/FlyCV/build/third_party/libjpeg-turbo/output/share/doc/libjpeg-turbo/wizard.txt
-- Installing: C:/_/3rd/FlyCV/build/third_party/libjpeg-turbo/output/share/doc/libjpeg-turbo/LICENSE.md
-- Installing: C:/_/3rd/FlyCV/build/third_party/libjpeg-turbo/output/lib/pkgconfig/libjpeg.pc
-- Installing: C:/_/3rd/FlyCV/build/third_party/libjpeg-turbo/output/lib/pkgconfig/libturbojpeg.pc
-- Installing: C:/_/3rd/FlyCV/build/third_party/libjpeg-turbo/output/lib/cmake/libjpeg-turbo/libjpeg-turboConfig.cmake
-- Installing: C:/_/3rd/FlyCV/build/third_party/libjpeg-turbo/output/lib/cmake/libjpeg-turbo/libjpeg-turboConfigVersion.cmake
-- Installing: C:/_/3rd/FlyCV/build/third_party/libjpeg-turbo/output/lib/cmake/libjpeg-turbo/libjpeg-turboTargets.cmake
-- Installing: C:/_/3rd/FlyCV/build/third_party/libjpeg-turbo/output/lib/cmake/libjpeg-turbo/libjpeg-turboTargets-release.cmake
-- Installing: C:/_/3rd/FlyCV/build/third_party/libjpeg-turbo/output/include/jconfig.h
-- Installing: C:/_/3rd/FlyCV/build/third_party/libjpeg-turbo/output/include/jerror.h
-- Installing: C:/_/3rd/FlyCV/build/third_party/libjpeg-turbo/output/include/jmorecfg.h
-- Installing: C:/_/3rd/FlyCV/build/third_party/libjpeg-turbo/output/include/jpeglib.h
[4/88] Performing build step for 'libpng'
FAILED: third_party/libpng/stamp/libpng-build C:/_/3rd/FlyCV/build/third_party/libpng/stamp/libpng-build
cmd.exe /C "cd /D C:\_\3rd\FlyCV\build\third_party\libpng && "C:\Program Files\CMake\bin\cmake.exe" --build . && "C:\Program Files\CMake\bin\cmake.exe" -E touch C:/_/3rd/FlyCV/build/third_party/libpng/stamp/libpng-build"
[1/24] Generating pnglibconf.out
FAILED: pnglibconf.out C:/_/3rd/FlyCV/build/third_party/libpng/pnglibconf.out
cmd.exe /C "cd /D C:\_\3rd\FlyCV\build\third_party\libpng && "C:\Program Files\CMake\bin\cmake.exe" -DINPUT=C:/_/3rd/FlyCV/build/third_party/libpng/pnglibconf.c -DOUTPUT=C:/_/3rd/FlyCV/build/third_party/libpng/pnglibconf.out -P C:/_/3rd/FlyCV/build/third_party/libpng/scripts/genout.cmake"
鐢ㄤ簬 x64 鐨?Microsoft (R) C/C++ 浼樺寲缂栬瘧鍣?19.34.31935 鐗?
鐗堟潈鎵€鏈?C) Microsoft Corporation銆備繚鐣欐墍鏈夋潈鍒┿€?

pnglibconf.c
C:/_/3rd/FlyCV/build/third_party/libpng/pnglibconf.c(34): fatal error C1083: 鏃犳硶鎵撳紑鍖呮嫭鏂囦欢: 鈥渮lib.h鈥? No such file or directory
CMake Error at scripts/genout.cmake:78 (message):
  Failed to generate
  C:/_/3rd/FlyCV/build/third_party/libpng/pnglibconf.out.tf1


ninja: build stopped: subcommand failed.
[5/88] Completed 'libjpeg-turbo'
ninja: build stopped: subcommand failed.

单独拎出来报错信息：

C:\_\3rd\FlyCV\build>cmd.exe /C "cd /D C:\_\3rd\FlyCV\build\third_party\libpng && "C:\Program Files\CMake\bin\cmake.exe" -DINPUT=C:/_/3rd/FlyCV/build/third_party/libpng/pnglibconf.c -DOUTPUT=C:/_/3rd/FlyCV/build/third_party/libpng/pnglibconf.out -P C:/_/3rd/FlyCV/build/third_party/libpng/scripts/genout.cmake"
用于 x64 的 Microsoft (R) C/C++ 优化编译器 19.34.31935 版
版权所有(C) Microsoft Corporation。保留所有权利。

pnglibconf.c
C:/_/3rd/FlyCV/build/third_party/libpng/pnglibconf.c(34): fatal error C1083: 无法打开包括文件: “zlib.h”: No such file or directory
CMake Error at scripts/genout.cmake:78 (message):
  Failed to generate
  C:/_/3rd/FlyCV/build/third_party/libpng/pnglibconf.out.tf1



C:\_\3rd\FlyCV\build>

armlinux编译报错

ARM LINUX 交叉编译报错，shuffle应改为：_shuffle，修改后正常

FlyCV-develop/modules/img_transform/color_convert/src/color_convert_arm.cpp:3515:26: error: ‘shuffle’ was not declare
d in this scope
:"r"(shuffle)

FlyCV-develop/modules/img_transform/color_convert/src/color_convert_arm.cpp:3515:26: note: suggested alternative: ‘_s
huffle’
:"r"(shuffle)
~~~~~~
_shuffle

画图函数失效

flycv版本：flycv-v1.0.0-android.tar
平台：展锐T610 CPU armv7
复现步骤：读取本地图片，然后调用画图函数，最后保存画图后的数据，保存的图片上没有画图结果

参考代码：
std::string raw_img_path = "/sdcard/Android/data/com.baidu.paddle.lite.demo.ssd_detection/cache/raw.jpg";
fcv::Mat raw_img = fcv::imread(raw_img_path);
fcv::circle(raw_img,fcv::Point(500,500),100,fcv::Scalar(0,0,100),2);

std::string res_img_path = "/sdcard/Android/data/com.baidu.paddle.lite.demo.ssd_detection/cache/raw_res.jpg";
fcv::imwrite(res_img_path, raw_img);

warp_affine函数与opencv接口不一致

怎么感觉在arm上加速不是很明显

下图是和opencv的resize接口耗时对比

用的开发板如下

IOS中flycv读取图片时报错：Rebuild fcv with libjpeg-turbo! 和Input Mat of resize is empty!

IOS中使用flycv和opencv的静态库,flycv读取图片时报错：Rebuild fcv with libjpeg-turbo! 和Input Mat of resize is empty!
flycv的版本：IOS flycv-v1.1.0-ios.tar
在flycv的库里查到了jpg的支持符号如下：

绘图方法

请问一下，flycv有没有类似于putText方法，在图像上绘制文字，并且支持中文的

什么时候python能用

什么时候python能用

直接 pip install FlyCV

这样就可以替换opencv

另外可以增加读取摄像头 读取视频文件的
图像处理 增加 裁剪

support cv::rectangle+cv::putText

cv::rectangle和cv::putText在AI cv里面使用频率非常高，是否可以考虑支持

建议提供只有C API导出的构建选项

目前如果使用-DBUILD_C=ON，可以构建C API的导出，但是同时也存在所有的C++ API的导出，这在利用C API的使用场景中（如C#或go语言封装、C API做其它功能等）是完全没有必要的，还会偏离FlyCV“小巧精致”、“高订制化”的初心。

我建议提供只有C API导出的构建选项（或者将-DBUILD_C=ON改为仅编译C API），在这个选项中，完全不需要C++的函数导出。

多线程稳定崩溃

开发者您好：

我在使用flycv代码替换opencv实现数据预处理函数时发现，若同时开多个推理线程（包括预处理、推理和后处理)，替换flycv后的代码在flycv线程池部分稳定发生崩溃，只有在预处理函数前后加入线程锁才能稳定运行。但opencv不论开几个线程都不会发生这样的现象。烦请排查。
还想请教一下目前imdecode是否无法读取大于4000像素的图片，提示不支持的格式。

armlinux交叉编译报错

/install/flycv/lib/libflycv_shared.so: undefined reference to `std::thread::_State::~_State()@GLIBCXX_3.4.22'
/install/flycv/lib/libflycv_shared.so: undefined reference to `typeinfo for std::thread::_State@GLIBCXX_3.4.22'
/install/flycv/lib/libflycv_shared.so: undefined reference to `std::thread::_M_start_thread(std::unique_ptr\<std::thread::_State, std::default_delete<std::thread::_State> >, void (*)())@GLIBCXX_3.4.22'

X86平台下编译报错

编译分支：v1.2.0
编译环境: X86 Linux
编译器:gcc (GCC) 7.4.0

使用指令：
cmake
-DBUILD_TEST=OFF
-DBUILD_BENCHMARK=OFF
-DCMAKE_INSTALL_PREFIX=${build_dir}/install
-DWITH_LIB_PNG=ON
-DWITH_LIB_JPEG_TURBO=ON
-DCMAKE_BUILD_TYPE=Release
-DENABLE_AVX=ON
..
make -j32
编译报错：
/world/FlyCV/modules/core/base/include/common_avx.h: In member function ‘void fcv::Vld3_F32x8_Avx::load(const float*, __m256i*, __m256i*, __m256i*)’:
/home/chenjunpin/world/FlyCV/modules/core/base/include/common_avx.h:366:25: error: ‘_mm256_loadu2_m128i’ was not declared in this scope
__m256i bgr_0 = _mm256_loadu2_m128i((__m128i const*)(src_f32 + 12), (__m128i const*)(src_f32 + 0));
^~~~~~~~~~~~~~~~~~~
/world/FlyCV/modules/core/base/include/common_avx.h:366:25: note: suggested alternative: ‘_mm256_loadu_si256’
__m256i bgr_0 = _mm256_loadu2_m128i((__m128i const*)(src_f32 + 12), (__m128i const*)(src_f32 + 0));

1、使用 gcc -march=native -E -v - </dev/null 2>&1 | grep avx 查看GCC支持,确认编译器支持-mavx -mavx2
2、请问是否是因为GCC版本不一致的原因导致报错？

ThreadPool线程个数

请教下为什么线程池仅设置线程个数为1或者2？难道不应该是cpu核数吗？
ThreadPool::ThreadPool() : _active_wait(10000) { _thread_num = get_cpu_num() > 1 ? 2 : 1; }

请问编译armhf需要什么依赖库呢

[ 19%] Completed 'zlib'
[ 19%] Built target zlib
[ 20%] Linking CXX static library libbenchmark.a
[ 20%] Built target benchmark
[ 21%] Linking CXX static library ../../../lib/libgtest.a
[ 21%] Built target gtest
make: *** [all] Error 2
[ 0%] Performing build step for 'libjpeg-turbo'
make[3]: *** No targets specified and no makefile found. Stop.
make[2]: *** [third_party/libjpeg-turbo/stamp/libjpeg-turbo-build] Error 2
make[1]: *** [CMakeFiles/libjpeg-turbo.dir/all] Error 2
make: *** [all] Error 2
出现了上面问题是要单独手动安装吗

Provide vcpkg build option

Please consider provide vcpkg build option: https://github.com/microsoft/vcpkg

JPG图像保存体积过大

相同的处理方式，生成的图片大小不一样。

希望所有的接口支持float类型，这是最基本的算法类型，然后才是U8,不然开源出来也用不了哈

量化计算请问如何组合现有函数比较好？

需要将输入数据进行量化计算，请问如何处理比较好？

add_subdirectory方式编译出错

Cmake中已add_subdirectory方式编译flycv，编译出错

编译错误

windows 目前只有支持vs 2015吗？

fcv::crop函数

• lycv版本：flycv-v1.0.0-android.tar
• 平台：展锐T610 CPU armv7
• 复现步骤：读取本地图片，然后在A函数中调用fcv::crop函数，最后在B函数中调用A函数，编译可以过，但是链接出错

#include "flycv.h"

void A(fcv::Mat& src_img,fcv::Mat& dst_img){
fcv::crop(src_img, dst_img, fcv::RectI(0,0,100,100));
}
int main(){
std::string raw_img_path = "/sdcard/Android/data/com.baidu.paddle.lite.demo.ssd_detection/cache/raw.jpg";
fcv::Mat raw_img = fcv::imread(raw_img_path);

fsv::dst_img;
A(raw_img,dst_img);

}

• 现象：注释掉A函数中的fcv::crop，能够成功编译运行或者在main函数中不调用也能成功编译运行
• 输出错误信息：
  error: undefined reference to 'fcv::crop(fcv::Mat&, fcv::Mat&, fcv::Rect_)'

希望添加一下常用的滤波方法。

高斯，均值滤波希望能添加一下。

当前如何实现对灰度图的normalize

normalize_to_submean_to_reorder只能对三通道图使用，看了下加减乘除或者convertTo暂时没有实现，请问目前有没有办法实现对灰度图的normalize呢？

CVT_PA_BGR2PA_RGB不生效

测试代码段
fcv::Mat dst;
fcv::Mat src = fcv::imread("user.jpg");
for (int i = 0; i < 20; ++i)
{
unsigned char* p = (unsigned char*)src.data();
int a = p[i];
std::cout << a << " ";
}
std::cout << std::endl;
std::cout << "**" <<std::endl;
fcv::cvt_color(src, src, fcv::ColorConvertType::CVT_PA_BGR2PA_RGB);
for (int i = 0; i < 20; ++i)
{
unsigned char p = (unsigned char)src.data();
int a = p[i];
std::cout << a << " ";
}

输出结果：
65 71 48 66 72 49 66 71 50 65 70 49 64 69 48 67 72 51 70 76

---

65 71 48 66 72 49 66 71 50 65 70 49 64 69 48 67 72 51 70 76

如何对接入参是cvmat的函数

项目中很多函数，包括第三方接口，入参需求是cvmat，我用flycv拿到的是fcvmat，该如何处理呢

有3个C API导出名称风格规范不一致

注意到有3个C API导出名称不一致：
分别是：

• get_cpu_num
• get_thread_num
• set_thread_num

应该使用整体的规范，比如：

• fcvGetCpuNum
• fcvGetThreadNum
• fcvSetThreadNum

建议更具体一些的名字，比如：

• fcvGetPhysicalThreadCount
• fcvGetLogicalThreadCount
• fcvSetThreadPoolThreadCount

希望FlyCV能提供C API以方便其它编程语言封装调用

FlyCV都是基于C++ API，希望能提供C API，以便我能使用C#/.NET等编程语言直接封装。

poly_lines 线条颜色怎么指定？

int poly_lines(Mat& img,
const Point2l* v,
int count,
bool is_closed,
const void* color,
int thickness,
LineType line_type,
int shift);
函数中颜色是const void*类型，这个函数是不是还没有开发完呢？

编译时第三方依赖libturbo-jpeg/libpng/zlib等在Windows环境无法下载

这几个第三方依赖无法自动下载，但可以手动下载，我觉得可以优化一下，怀疑可能和这下bash有关：

这个问题可以通过手动下载第三方依赖来避过去：

cd third_party
git clone --progress --depth=1 -b 2.1.4 [email protected]:libjpeg-turbo/libjpeg-turbo.git libjpeg-turbo
git clone --progress --depth=1 -b libpng16 [email protected]:glennrp/libpng.git libpng
git clone --progress --depth=1 -b v1.2.9 [email protected]:madler/zlib.git zlib
git clone --progress --depth=1 -b release-1.12.1 [email protected]:google/googletest.git gtest
git clone --progress --depth=1 -b v1.6.1 [email protected]:google/benchmark.git benchmark

建议优化fcv_download_dependency，避免在Windows环境有一些不太方便或明确的依赖。

fcv::imread解码png图片

平台ARM64 版本V1.1.0

1.1.0文档中说明了解图仅支持JPEG；
链接libpng，发现png解出来的是RGB。。。是不是加转下BGR比较好？

fcv::imwrite("xxx.jpg", mat)时图片过大

平台架构：ARM64
版本：v1.1.0

预编译版和自行编译都试过了。
发现写JPEG图片的时候，体积非常大...比raw data还大？

JPEG filesize=16,812,032 bytes
raw data size=2048*1365*3=8,386,560 bytes

#include <flycv.h>

#include <fstream>
#include <iostream>
#include <opencv2/opencv.hpp>
#include <string>

int main(int argc, char **argv) {
  if (argc < 2) {
    fprintf(stderr, "Usage: %s <img-path>\n", argv[0]);
    exit(1);
  }
  std::string image_path = argv[1];

  cv::Mat cv_img = cv::imread(image_path);
  int width = cv_img.cols;
  int height = cv_img.rows;
  int channels = cv_img.channels();
  printf("image width=%d, height=%d, channels=%d\n", width, height, channels);
  cv::imwrite("cv.jpg", cv_img);

  fcv::Mat fcv_img(width, height, fcv::FCVImageType::PKG_BGR_U8);
  memcpy(fcv_img.data(), cv_img.data, width * height * channels * sizeof(uint8_t));
  fcv::imwrite("fcv.jpg", fcv_img);

  return 0;
}

截图：

所用图片：

windows10 vs2022编译报错

又试了一次，直接报错

copymakeborder新增其他模式

希望能尽快增加copymakeborder的其他模式，谢谢！

希望尽快完善Mat的加减乘除运算，及conver_to等这些最基本使用率最高的算子。

flycv 支持 vs插件ImageWatch吗?

请问flycv 支持 vs插件ImageWatch吗,能支持这个很重要,调试方便? 或者,opencv 的mat 可以和flycv的mat互相转换也行.

cvt_color 转换格式 bgr2yuv

平台: ARM64
版本: release/v1.1.0

 fcv::Mat fcv_img = fcv::imread(image_path);
  printf("image width=%d, height=%d, channels=%d\n", fcv_img.width(), fcv_img.height(),
         fcv_img.channels());

  int padding_h = (fcv_img.height() + 1) / 2 * 2 - fcv_img.height();
  int padding_w = (fcv_img.width() + 1) / 2 * 2 - fcv_img.width();
  fcv::copy_make_border(fcv_img, fcv_img, 0, padding_h, 0, padding_w,
                        fcv::BorderType::BORDER_CONSTANT, fcv::Scalar(114, 114, 114));

  fcv::Mat fcv_yuv_img;
  fcv::cvt_color(fcv_img, fcv_yuv_img, fcv::ColorConvertType::CVT_PA_BGR2NV21);
  printf("yuv image width=%d, height=%d, channels=%d\n", fcv_yuv_img.width(), fcv_yuv_img.height(),
         fcv_yuv_img.channels());

跑到 padding 的时候会崩掉：gdb

Using host libthread_db library "/lib/aarch64-linux-gnu/libthread_db.so.1".
image width=2048, height=1365, channels=3

Program received signal SIGSEGV, Segmentation fault.
__memcpy_generic () at ../sysdeps/aarch64/multiarch/../memcpy.S:185
185     ../sysdeps/aarch64/multiarch/../memcpy.S: No such file or directory.
(gdb) bt
#0  __memcpy_generic () at ../sysdeps/aarch64/multiarch/../memcpy.S:185
#1  0x0000000000415b58 in fcv::copy_make_border_common(fcv::Mat&, fcv::Mat&, int, int, int, int, fcv::BorderType, fcv::Scalar_<double> const&)
    ()
#2  0x0000000000414394 in fcv::copy_make_border(fcv::Mat&, fcv::Mat&, int, int, int, int, fcv::BorderType, fcv::Scalar_<double> const&) ()
#3  0x0000000000405dec in MiniPadding (img=...) at /home/work/AtlasProjects/MaiAtlas/flycv_test/convert_test.cpp:23
#4  0x0000000000405f18 in main (argc=3, argv=0xfffffffff038) at /home/work/AtlasProjects/MaiAtlas/flycv_test/convert_test.cpp:40

另外，有时候会崩在cvt_color处
w=2048, h=1365
padding成功
w=2048, h=1366

Thread 1 "test" received signal SIGSEGV, Segmentation fault.
_int_malloc (av=av@entry=0xfffeff8e3a70 <main_arena>, bytes=bytes@entry=638400) at malloc.c:3924
3924    malloc.c: No such file or directory.
(gdb) bt
#0  _int_malloc (av=av@entry=0xfffeff8e3a70 <main_arena>, bytes=bytes@entry=638400) at malloc.c:3924
#1  0x0000fffeff807444 in __GI___libc_malloc (bytes=638400) at malloc.c:3075
#2  0x000000000065e720 in fcv::cpu_allocator::cpu_allocator(unsigned long) ()
#3  0x000000000065e284 in fcv::get_allocator_from_platform(unsigned long, fcv::PlatformType, int) ()
#4  0x000000000064ce68 in fcv::Mat::Mat(int, int, fcv::FCVImageType, int, fcv::PlatformType) ()
#5  0x000000000064d7ac in fcv::cvt_color(fcv::Mat const&, fcv::Mat&, fcv::ColorConvertType) ()

有没有文档图片矫正的功能

怎样用这个实现文档图片矫正的功能

QNX交叉编译报错

QNX交叉编译报错
imgcodecs.cpp 函数 strcasecmp未定义，需添加<string.h>