statically compiling opencv

[rhobro]

Hi there,

I saw on your README that opencv has to be installed on the device for it to run. I am writing an app in flutter (aimed at android, ios, windows, linux, macos) which has bits of rust using opencv. How can I package opencv with the app? This is mostly for the mobile versions which may not have it installed.

Thanks

[twistedfall]

Probably static linking is a solution, I haven’t tested this option, but there were a couple of successful mentions. And working with mobile devices is also something I haven’t really considered so far. So if you manage to get something working I would really like to hear about it!

[rhobro]

Hi there,

I read on the opencv website that the library was adjusted for ios and android. So I think compiling them for the platforms should work?

I am new to rust bindings to C and C++. I'm trying to dig through the build script. How does one go about statically linking opencv?

Thanks @twistedfall

[rhobro]

Would it be possible to introduce a feature flag on the crate to decide between static and dynamic linking?

[twistedfall]

So I think compiling them for the platforms should work?

In theory, yes, but I can't say that I tried that myself

How does one go about statically linking opencv? Would it be possible to introduce a feature flag on the crate to decide between static and dynamic linking?

In theory it's just about the linker flags. If you manage to get something working I would really appreciate the PR. Having a feature flag for switching between dynamic and static linking is a good idea!

[rhobro]

I've forked the repo and have edited the RUSTFLAGS. I've been trying to test it out in a local cargo project but it seems to return errors saying that all the probes failed to find opencv.

It is returning errors that environment and vcpkg cannot probe it properly - how can I fix this in the build script?

[rhobro]

It seems to work when I build my program for my host but it seems to break when cross-compiling. It says environment and vcpkg cannot probe opencv.

[spoorn]

Probably static linking is a solution, I haven’t tested this option, but there were a couple of successful mentions. And working with mobile devices is also something I haven’t really considered so far. So if you manage to get something working I would really like to hear about it!

Are there any examples of how to do the static linking?

[rhobro]

I read that changing the LDFLAGS can switch between dynamic and static. But it seems to not include opencv in the dylib.

[twistedfall]

I'm going to close this issue in favor of the already existing one about static compilation: https://github.com/twistedfall/opencv-rust/issues/104

[jb-alvarado]

Hello, I don't know if it fits here, but I wanted to share my steps, to build with static openCV libs. I was not able to use as target x86_64-unknown-linux-musl, I think for that openCV has to compile also with a musl compiler.

In my project I only use features: face and imgproc: opencv = { version = "0.73", default-features = false, features = ["face", "imgproc"]}

Here are the compiling steps:

1. download sources:

   •   wget -O opencv.zip https://github.com/opencv/opencv/archive/4.x.zip
       wget -O opencv_contrib.zip https://github.com/opencv/opencv_contrib/archive/4.x.zip
     

2. unzip sources, and delete archive:

   •   unzip opencv.zip && rm opencv.zip
       unzip opencv_contrib.zip && rm opencv_contrib.zip
     

3. create build folder and switch to it:

   •   mkdir -p build && cd build
     

4. configure build

   •   cmake -DCMAKE_BUILD_TYPE=Release -DBUILD_SHARED_LIBS=NO \
       -DCMAKE_INSTALL_PREFIX=/opt/opencv -DBUILD_DOCS=OFF -DBUILD_EXAMPLES=OFF \
       -DBUILD_TESTS=OFF -DBUILD_PERF_TESTS=OFF -DBUILD_opencv_java=OFF \
       -DBUILD_opencv_python=OFF -DOPENCV_EXTRA_MODULES_PATH=../opencv_contrib-4.x/modules \
       ../opencv-4.x
     

5. compile and install opencv

   •   make -j8 && make install
     

6. go to rust project
7. compile rust project with:

   •   OPENCV_LINK_LIBS="opencv_imgproc,opencv_face,opencv_objdetect,opencv_dnn,opencv_dnn_objdetect,opencv_core,ippiw,ittnotify,ippicv,liblibprotobuf,z" \
       OPENCV_LINK_PATHS=/opt/opencv/lib,/opt/opencv/lib/opencv4/3rdparty,/usr/lib/x86_64-linux-gnu \
       OPENCV_INCLUDE_PATHS=/opt/opencv/include/opencv4 \
       cargo build --release
     

[twistedfall]

That's really helpful, thank you! I will try to include this step into the CI and update the readme

[jb-alvarado]

I will try to include this step into the CI and update the readme

Maybe you should also mention that it is a bit tricky to get the right ordering from libs in OPENCV_LINK_LIBS, for example when you put libippicv before libippiw the compiler doesn't find all dependencies.

With libz I was not sure which one to choice, the static or the dynamic one. The rust program will link against the dynamic library, but openCV I guess will link to the static one.

[twistedfall]

I've added the static build to CI and updated the readme: https://github.com/twistedfall/opencv-rust#static-build

[00hello]

I've added the static build to CI and updated the readme: https://github.com/twistedfall/opencv-rust#static-build

@twistedfall

Question: I'm trying to statically compile opencv, as well (on Ubuntu focal fossa). I'm not sure which script I'm supposed to run, install-focal.sh or script.sh. Why do you have a linux install script in both; what does one do that the other doesn't?

And if I do run install-focal.sh, how do I run it? Do I run ./install-focal.sh directly from command line? or do I run ./install.sh (install.sh, which seems to call install-focal.sh) with $OS_FAMILY environment variable set to Linux?

[twistedfall]

For installation of the static OpenCV itself on Ubuntu use the install-focal.sh (Focal stands for Ubuntu Focal Fossa, 20.04 version), just be sure to set up env var OPENCV_VERSION=4.7.0-static before doing that. The build will be done in $HOME/build/opencv/ directory and installed to /usr (you might want to change that to /usr/local or something like this by editing the script before running).

[spoorn]

For anyone looking at this in the future, I have a github actions workflow file that statically links OpenCV into my binary for Windows, macOS, and Linux

There are also comments in workflow file to help understand the differences between platforms