add support for desktop examples on MacOS (Apple Silicon M1=arm64/ x86_64/ i386)

[walkoncross]

1. add config "build:darwin_arm64" in .bazelrc;
2. add "Build on MacOS" section in README.rd;
3. add ‘config_setting(name = "macos_arm64", …)’ in mediapipe/BUILD, and set default apple_platform_type
   to ‘macos_x86_64’;
4. add build_macos_desktop_examples.sh 

[google-cla]

Thanks for your pull request. It looks like this may be your first contribution to a Google open source project (if not, look below for help). Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

📝 Please visit https://cla.developers.google.com/ to sign.

Once you've signed (or fixed any issues), please reply here with @googlebot I signed it! and we'll verify it.

---

What to do if you already signed the CLA

Individual signers

• It's possible we don't have your GitHub username or you're using a different email address on your commit. Check your existing CLA data and verify that your email is set on your git commits.

Corporate signers

• Your company has a Point of Contact who decides which employees are authorized to participate. Ask your POC to be added to the group of authorized contributors. If you don't know who your Point of Contact is, direct the Google project maintainer to go/cla#troubleshoot (Public version).
• The email used to register you as an authorized contributor must be the email used for the Git commit. Check your existing CLA data and verify that your email is set on your git commits.
• The email used to register you as an authorized contributor must also be attached to your GitHub account.

ℹ️ Googlers: Go here for more info.

[walkoncross]

@googlebot I signed it!

[mgebhard]

Super helpful I finally got helloworld to build on my m1 but any help with:

In file included from mediapipe/util/annotation_renderer.cc:15:
In file included from ./mediapipe/util/annotation_renderer.h:20:
./mediapipe/framework/port/opencv_core_inc.h:18:10: fatal error: 'opencv2/core/version.hpp' file not found
#include <opencv2/core/version.hpp>
^~~~~~~~~~~~~~~~~~~~~~~~~~
1 error generated.
Error in child process '/usr/bin/xcrun'. 1
Target //mediapipe/examples/desktop/face_detection:face_detection_cpu failed to build
Use --verbose_failures to see the command lines of failed build steps.
INFO: Elapsed time: 1.591s, Critical Path: 1.41s
INFO: 175 processes: 131 internal, 44 darwin-sandbox.

[terasum]

@mgebhard try modifiy the third_party/opencv_macos.BUILD :

--- a/third_party/opencv_macos.BUILD
+++ b/third_party/opencv_macos.BUILD
@@ -19,8 +19,8 @@ cc_library(
             "lib/libopencv_videoio.dylib",
         ],
     ),
-    hdrs = glob(["include/opencv2/**/*.h*"]),
-    includes = ["include/"],
+    hdrs = glob(["include/opencv4/opencv2/**/*.h*"]),
+    includes = ["include/opencv4/"],
     linkstatic = 1,
     visibility = ["//visibility:public"],
 )

[a-hurst]

@walkoncross Thanks a ton for this! Hopefully it gets merged soon.

NOTE: My original solution doesn't work because Bazel doesn't allow selects for repository paths, but the following solution works:

In WORKSPACE replace the macOS lines with these:

new_local_repository(
    name = "macos_opencv",
    build_file = "@//third_party:opencv_macos.BUILD",
    path = "/usr/local/opt/opencv@3",
)

new_local_repository(
    name = "macos_arm64_opencv",
    build_file = "@//third_party:opencv_macos.BUILD",
    path = "/opt/homebrew/opt/opencv@3",
)

new_local_repository(
    name = "macos_ffmpeg",
    build_file = "@//third_party:ffmpeg_macos.BUILD",
    path = "/usr/local/opt/ffmpeg@4",
)

new_local_repository(
    name = "macos_arm64_ffmpeg",
    build_file = "@//third_party:ffmpeg_macos.BUILD",
    path = "/opt/homebrew/opt/ffmpeg@4",
)

Then, in third_party/BUILD, add the platform-specific library paths for each macOS architecture like so:

alias(
    name = "opencv_binary",
    actual = select({
        "//mediapipe:android_x86": "@android_opencv//:libopencv_x86",
        "//mediapipe:android_x86_64": "@android_opencv//:libopencv_x86_64",
        "//mediapipe:android_armeabi": "@android_opencv//:libopencv_armeabi-v7a",
        "//mediapipe:android_arm": "@android_opencv//:libopencv_armeabi-v7a",
        "//mediapipe:android_arm64": "@android_opencv//:libopencv_arm64-v8a",
        "//mediapipe:ios": "@ios_opencv//:opencv",
        "//mediapipe:macos_x86_64": "@macos_opencv//:opencv",
        "//mediapipe:macos_arm64": "@macos_arm64_opencv//:opencv",
        "//mediapipe:windows": "@windows_opencv//:opencv",
        "//conditions:default": "@linux_opencv//:opencv",
    }),
)

cc_library(
    name = "libffmpeg",
    visibility = ["//visibility:public"],
    deps = select({
        "//mediapipe:android_x86": [],
        "//mediapipe:android_x86_64": [],
        "//mediapipe:android_armeabi": [],
        "//mediapipe:android_arm": [],
        "//mediapipe:android_arm64": [],
        "//mediapipe:ios": [],
        "//mediapipe:macos_x86_64": [
            "@macos_ffmpeg//:libffmpeg",
        ],
        "//mediapipe:macos_arm64": [
            "@macos_arm64_ffmpeg//:libffmpeg",
        ],
        "//conditions:default": [
            "@linux_ffmpeg//:libffmpeg",
        ],
    }),
)

Anyway, that lets you build easily on x86 or arm64 macOS without having to edit any build files, which should hopefully make it easier for the project to provide official macOS arm64 wheels.

That said, with the changes in this PR (and my edits above) I can install mediapipe with python setup.py install --link-opencv in a clean virtual environment just fine, but python setup.py bdist_wheel fails with an error about missing symbols when compiling OpenCV:

...
Scanning dependencies of target opencv_annotation
[ 99%] Building CXX object apps/annotation/CMakeFiles/opencv_annotation.dir/opencv_annotation.cpp.o
[ 99%] Linking CXX executable ../../bin/opencv_annotation
Undefined symbols for architecture arm64:
  "_jpeg_default_qtables", referenced from:
      cv::JpegEncoder::write(cv::Mat const&, std::__1::vector<int, std::__1::allocator<int> > const&) in libopencv_imgcodecs.a(grfmt_jpeg.cpp.o)
ld: symbol(s) not found for architecture arm64
clang: error: linker command failed with exit code 1 (use -v to see invocation)
Error in child process '/usr/bin/xcrun'. 1
/private/var/tmp/_bazel_austinhurst/c5b55272d6d862676b481c1414e3a50d/sandbox/darwin-sandbox/1778/execroot/mediapipe/external/local_config_cc/cc_wrapper.sh: line 69: 77421 Abort trap: 6           "$(/usr/bin/dirname "$0")"/wrapped_clang "$@"
make[2]: *** [bin/opencv_annotation] Error 134
make[1]: *** [apps/annotation/CMakeFiles/opencv_annotation.dir/all] Error 2

If I update the OpenCV source version to 3.4.16 I can get past it (see this OpenCV issue: opencv/opencv#18541), but then I run into the same bizarre ld: library not found for -ljpeg error reported in issue #1665.

[cansik]

Would be great if mediapipe releases M1 packages officially. I managed to create a fully automated script to build version 0.8.9 with python wheels. I link opencv@3 and later delocate the wheel paths. The prebuilt wheels for version 3.8, 3.9 and 3.10 are already uploaded and after some testing I will release them on PyPi as well. Looking forward to feedback and suggestions.

Here the repository with the build: https://github.com/cansik/mediapipe-silicon

[a-hurst]

Would be great if mediapipe releases M1 packages officially. I managed to create a fully automated script to build version 0.8.9 with python wheels. I link opencv@3 and later delocate the wheel paths. The prebuilt wheels for version 3.8, 3.9 and 3.10 are already uploaded and after some testing I will release them on PyPi as well. Looking forward to feedback and suggestions.

Here the repository with the build: https://github.com/cansik/mediapipe-silicon

Fantastic, thanks for putting this together! Will vastly simplify the setup and install process for anything that depends on mediapipe. For my use-case (extracting the x/y coordinates of face regions for cognitive face processing research), I can't realistically expect colleagues to set up and troubleshoot a compiler environment just to run a script, so having pre-built wheels is extremely useful.

[schmidt-sebastian]

Thank you for your contribution! We have integrated the required changes from your PR. We now support M1 for iOS and Android builds (Python support is available via Rosetta).