r0ml/opencv

Every time I try to use OpenCV on an Apple platform, I lose a few days to yak shaving trying to

Building your own version

First, if you wish to update the underlying version of OpenCV, cd into opencv and do: git fetch --all --tags git checkout <tagname>

then do the same in opencv_contrib (use the same tagname)

Then, back in the root directory of the package, git add opencv git add opencv_contrib git commit -am 'upgraded to opencv <tagname>'

If you wish to build your own, clone this repo (git clone --recurse-submodules). <br/> You will need cmake installed (brew install cmake). <br/> You will need XCode command line tools installed (xcode-select --install). <br/> You will need to create a version of python called python installed at /usr/local/bin . <br/> (The build scripts want an executable called python in the PATH. Unfortunately, if you create a link to /usr/bin/python3, it will not work! )

I have included a script called python in this repo which must be copied or moved to /usr/local/bin/python.

Then run

./prepatch.sh

which fixes a problem with an #include statement in opencv and bad logic for macCatalyst support in TargetConditionals.h .

Then run

./build.sh

Because I am using OpenCV as the name of the framework, rather than opencv2, many of the includes in the headers fail to resolve properly. The following patch fixes that, as well as changes to XCode 26 in validating iOS frameworks.

Run

./postpatch.sh

The built xcframework should be located at build_xcframework/OpenCV.xcframework.

Deploying

After building the xcframework, one needs to 1) run ./postbuild.sh xxx where xxx is the new version number of the updated package. This will update the version number in the download URL, create the release zip file, and update the checksum in Package.swift 2) commit and push the changes 3) create a new release on github using the version number from the postbuild.sh step 4) upload the OpenCV.framework.zip in build_xcframework as the asset in the release

Alternate

If adding this SPM package seems to take way too long, you might try using the alternative package at https://github.com/r0ml/JustTheBinary. This contains a binary package with the same framework as available here, except (as the name suggests) it does not include all the stuff required to rebuild the framework. It should load faster.

LICENSE

Package Metadata

Repository: r0ml/opencv

Default branch: main

README: README.md