Skip to content

Latest commit

 

History

History
82 lines (65 loc) · 3.65 KB

MAC.md

File metadata and controls

82 lines (65 loc) · 3.65 KB
Raw

Install pyslam under macOS

The following procedure has been tested under Big Sur 11.2 and Xcode 12.0.

Please, follow these install steps:

  1. first you need to install homebrew and Xcode CLT (check the section below)
  2. download this repo and move to the experimental mac branch:
   $ git clone https://github.com/luigifreda/pyslam.git 
   $ cd pyslam 
   $ git checkout mac
  1. change your default shell type to bash: 
    $ chsh -s /bin/bash
    (if you want to set zsh back then run: $ chsh -s /bin/zsh)
  2. launch the macOS install script
   $ ./install_all_mac_venv.sh
  1. in order to run main_vo.py run
   $ ./launch_main_vo.sh
  1. in order to run main_slam.py run
   $ ./launch_main_slam.sh

NOTE 1: the above procedure will install a virtual python environment pyslam in your system. That virtual environment can be easily activated by using the command:

$ . pyenv-activate.sh

(do not forget the dot! without '/' ! ) You can find further details about python virtual environments here.

NOTE 2: the launch scripts above launch_main_xxx.sh will automatically activate the pyslam virtual enviroment for you and launch the scripts with the necessary environment variable setting (explained below):

$ OBJC_DISABLE_INITIALIZE_FORK_SAFETY=YES python3 main_xxx.py

NOTE 3: In order to make things running under macOS, I had to use some tricks (for matplotlib processes in particular, further details below). Please, consider that pyslam has been designed under Linux (Ubuntu 18.04), where you can get it in its 'best shape'.

Notes about macOS

Install homebrew

from https://brew.sh/

$ /usr/bin/ruby -e "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/master/install)"

You may also need to install gcc by using XCode Command Line Tools. Run

$ xcode-select --install

Issues found with matlplotlib

I found the following problems with python multi-processing (see https://stackoverflow.com/questions/50168647/multiprocessing-causes-python-to-crash-and-gives-an-error-may-have-been-in-progr). The proposed solution to run this command in the open shell

$ export OBJC_DISABLE_INITIALIZE_FORK_SAFETY=YES

does not work. In another thread https://stackoverflow.com/questions/50168647/multiprocessing-causes-python-to-crash-and-gives-an-error-may-have-been-in-progr#comments-52230415, I found something that does work with both pangolin processes and mplot processes: launch the main scripts by setting the same environment variable one the same line

$ OBJC_DISABLE_INITIALIZE_FORK_SAFETY=YES python3 xxx.py

I found other issues with matplotlib due to plt.ion() (interactive mode) that does not work on mac. In order to make the matplotlib processes working, I add to apply some other tricks that make the matplot figures being refreshed in an inelegant way (being activated and refreshed in turn one over the other). But it works! :-)
At the present time, pyslam is still experimental on macOS!

Issues found with OpenCV and pyenv

When you launch one of the scripts above, you get a warning:

objc[6169]: Class CaptureDelegate is implemented in both /Users/luigi/.python/venvs/pyslam/lib/python3.7/site-packages/cv2/cv2.cpython-37m-darwin.so (0x11923d590) and /usr/local/opt/opencv/lib/libopencv_videoio.4.3.dylib (0x13021d0c8). One of the two will be used. Which one is undefined.

This is an open issue which needs to be solved. In a few words, this is an "interference" between the OpenCV libs of the installed virtual python environment and the OpenCV libs installed by homebrew.