Running DepthAI on Apple M1 Based Macs

Brandon

So we got our Mac mini M1, and gave it a quick try running DepthAI on it. Got our prebuilt wheels (for Intel CPUs - x86_64) working very well using the Rosetta emulator.

First, the native Apple Silicon support for Homebrew is still being actively worked on (with several issues still existing), so I didn't attempted installing it / building from sources. See https://github.com/Homebrew/brew/issues/7857

The steps followed:

Install Rosetta (note it may be already installed, for example attempting to install other apps (current TeamViewer) may prompt to install Rosetta first):
softwareupdate --install-rosetta
To run commands with x86_64 emulation, it is possible to prefix them with arch -x86_64

But there's an easier way, to open the terminal app directly using Rosetta. For example, I copied Terminal to Desktop, renamed it to Terminal-x86_64 , and tick the checkbox Open using Rosetta:

Then after opening the terminal, it can be checked under what architecture it runs using uname -m

with Rosetta it prints: x86_64
native: arm64

After that, it's just the standard procedure to install Brew (it installed python 3.8.2 by default), clone depthai repo, running ./install_requirements.py , and finally ./depthai_demo.py.

Cheers,
The Luxonis Team

smd58tex

I was able to follow the instructions here for Mac with no issues on my M1 Macbook Air, however I used iterm2, and not terminal. I also did not run it in rosetta mode.

I am, however, having issues with connectivity on both a USBA-> C cable that came with the camera with an additional USBA->C adapter, and a short USBC/Thunderbolt cable (which I would assume would work)

Running with --force_usb2 works for now.

Brandon

smd58tex Thanks for following up!

RobotRoss

With a new Oak-D-Lite on my Mac Mini M1 Monterey, using a 2nd Rosetta enabled Terminal that shows "uname -m" is x86_64, I was unable to run "python3 -m pip install depthai", apparently due to conflict with the "default" installation in /opt/homebrew. Collecting depthai
Using cached depthai-2.14.0.0.tar.gz (325 kB)
Installing build dependencies ... done
Getting requirements to build wheel ... done
Preparing metadata (pyproject.toml) ... done Building wheels for collected packages: depthai
Building wheel for depthai (pyproject.toml) ... error
ERROR: Command errored out with exit status 1:
command: /opt/homebrew/opt/python@3.9/bin/python3.9 /opt/homebrew/lib/python3.9/site-packages/pip/vendor/pep517/in_process/in_process.py build_wheel /var/folders/_j/1spktm_s02xgtcbwv77rwfn80000gn/T/tmplmufspx7 ....and lots more errors !

erik

Hello RobotRoss , what do you mean by 2nd Rosetta enabled terminal? We should have prebuilt wheels for that (x86_64), so I am not sure why it's trying to build new wheels for that. Which python version are you using?
Thanks, Erik

RobotRoss

Erik, Tnx for response. My (Homebrew) installed python3 from arm64 terminal.app (uname -m arm6) is:
which python3
/opt/homebrew/bin/python3
Ross@Kepler-2 ~ % python3 --version
Python 3.9.9
From x86_64 terminal-rosetta2.app
which python
/usr/bin/python
Ross@Kepler-2 ~ % python3 --version
Python 3.9.9

I have a 1st native “terminal.app”, uname -m arm64. Homebrew is installed in /opt/homebrew & there is text in the PATH environment to this folder. This is the default, per this response to 'brew config" from the Native arm64 and Rosetta x86_64 terminal.
From arm64 terminal.app
Ross@Kepler-2 ~ % brew doctor
Your system is ready to brew.
Ross@Kepler-2 ~ % brew config
HOMEBREW_VERSION: 3.3.9
ORIGIN: https://github.com/Homebrew/brew
HEAD: 96137bc19e68398ebbb7033379df288cd8b9a3f9
Last commit: 4 weeks ago
Core tap ORIGIN: https://github.com/Homebrew/homebrew-core
Core tap HEAD: 551bdc22ddc5695e9db3e7a59c30b4dba83fc904
Core tap last commit: 23 hours ago
Core tap branch: master
HOMEBREW_PREFIX: /opt/homebrew
HOMEBREW_CASK_OPTS: []
HOMEBREW_CORE_GIT_REMOTE: https://github.com/Homebrew/homebrew-core
HOMEBREW_MAKE_JOBS: 8
Homebrew Ruby: 2.6.8 => /System/Library/Frameworks/Ruby.framework/Versions/2.6/usr/bin/ruby
CPU: octa-core 64-bit arm_firestorm_icestorm
Clang: 13.0.0 build 1300
Git: 2.32.0 => /Library/Developer/CommandLineTools/usr/bin/git
Curl: 7.77.0 => /usr/bin/curl
macOS: 12.1-arm64
CLT: 13.2.0.0.1.1638488800
Xcode: 13.2.1
Rosetta 2: false

From x86_64 terminal-rosetta2.app
Ross@Kepler-2 ~ % brew config
HOMEBREW_VERSION: 3.3.10
ORIGIN: https://github.com/Homebrew/brew
HEAD: 385892f3d27dcab0f5a1cb47c3927d9d65ed007c
Last commit: 33 hours ago
Core tap ORIGIN: https://github.com/Homebrew/homebrew-core
Core tap HEAD: aa2f643fd4215653c61d968ca54ff5c56b6f3da6
Core tap last commit: 20 hours ago
Core tap branch: master
HOMEBREW_PREFIX: /opt/homebrew
HOMEBREW_REPOSITORY: /opt/homebrew
HOMEBREW_CELLAR: /opt/homebrew/Cellar
HOMEBREW_CASK_OPTS: []
HOMEBREW_CORE_GIT_REMOTE: https://github.com/Homebrew/homebrew-core
HOMEBREW_MAKE_JOBS: 8
Homebrew Ruby: 2.6.8 => /System/Library/Frameworks/Ruby.framework/Versions/2.6/usr/bin/ruby
CPU: octa-core 64-bit westmere
Clang: 13.0.0 build 1300
Git: 2.34.1 => /opt/homebrew/bin/git
Curl: 7.77.0 => /usr/bin/curl
macOS: 12.1-x86_64
CLT: 13.2.0.0.1.1638488800
Xcode: 13.2.1
Rosetta 2: true
Ross@Kepler-2 ~ % uname -m
x86_64

The 2nd terminal-rosetta.app was configured enabled for Rosetta2, as described by your Brandon Dec 20 note “69-depthai-apple-m1-based-macs”. uname -m x86_64. There is a 2nd homebrew installed in the pre-M1 folder /use/local/homebrew This text is not in the PATH. I ran the python3 -m install_requirements.py from this Rosetta2 enabled terminal, with the earlier posted result. The system seems to want to run the arm64 version of Homebrew Python to build your depth wheels, even while using the x86_64 terminal, probably because of the PATH text , with the resulting errors. Is there a way to force the system to stay with the x86_64 terminal Homebrew for running python3 ?

So since the Mac M1 has been out for over a year now and presumably Homebrew has matured, can you release a native arm64 depthai wheel? I’d rather not potentially break my current Homebrew jury rigging the x86_64 Rosetta2 terminal. I note there are other Data Scientist packages in arm64 Homebrew , such as numpy, scipy.
Regards
RobotRoss

pierreia

To all users of M1 Mac that would like to use DepthAI, I recommend you to build a VM with Ubuntu ARM. My OAK-D works like a charm on it!

RobotRoss

Yes, I can confirm the same success with a Ubuntu 20.04 VM on my Mac M1 and that the scripts run fast. As described for a Linux installation, must enable the udev rule. I will look forward, though for release of a native MAC arm64 Homebrew wheel installation. I’ve also installed the depthai on a Jetson Nano B01 Ubuntu 18.04 and Raspberry Pi 4/4G Raspi OS Buster 32 bit arm7.( though the compile takes 35 mins)

erik

Hello RobotRoss , looking at piwheels, we should have 2.13.2 pre-built wheels for arm6/arm7 (python 3.7/3.9), as on the img below.

Thanks, Erik

AlSantana

Hi everyone!!
Yesterday I received my flaming new OAK-D Lite, and after 30 minutes my experience is not that D-Litefull ;-). And I haven't even plugged it in 😅 . I've followed Brandon instructions and copied a Terminal app on the desktop, selected Open using Rosetta, checked that the uname -m was X84_64, and still get the following error (exactly the same I had before re-creating the "Rosetta Terminal"):

Error: Cannot install under Rosetta 2 in ARM default prefix (/opt/homebrew)!
To rerun under ARM use:
arch -arm64 brew install ...
To install under x86_64, install Homebrew into /usr/local.

I have checked that there is a Homebrew folder in /usr/local, but I guess the code is using the native Homebrew installation. Any ideas or suggestions

EDIT:
I have fixed this with:
export PATH=/usr/local/Homebrew/bin:$PATH in case it helps someone.

RobotRoss

If the Terminal is enabled for x86_64 ("Use with Rosetta" box checked" and there is NO previous Homebrew installed in the native arm64 Mac, this may work. The problem I had was that I had apreviously installed Homebrew which goes into /opt/homebrew space and sets "HOMEBREW_PREFIX: /opt/homebrew " and matching "PATH" locations. This apparently confuses the depthai installation scripts which installs another homebrew stack in "/usr/local/homebrew" space but does not completely configure the default paths . Execute "_% brew config" to see your default path settings. While I have NOT yet followed these instructions, check out the #macos for posts by whab which uses a virtual environment: https://discord.com/channels/790680891252932659/924798973519216651/934391777811636294 .

Orbitus007

This is not the answer.. compile the wheel for the M1 is the solution:

whab — 01/22/2022
For those struggling with running DepthAI natively on their M1 computers, here are step-by-step instructions to build, install and test a M1 native version of DepthAI in less than 30 minutes (tested on a new MacBookPro M1 Pro running macOS Monterey 12.1 with a OAK-D-Lite):

# Install native M1 version of brew
/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/master/install.sh)"
echo 'eval "$(/opt/homebrew/bin/brew shellenv)"' >> ~/.zprofile
eval "$(/opt/homebrew/bin/brew shellenv)"

# Install conda to create virtual environments for Python
brew install --cask miniconda
conda init zsh
Close and re-open a Terminal window

# Install DepthAI by building a M1 wheel (inside ~/DepthAI/)
conda create --name DepthAIEnv39 python=3.9
conda activate DepthAIEnv39
python3 -m pip install -U pip
brew update
brew install cmake libusb
cd ~; mkdir DepthAI; cd DepthAI
git clone --recursive  https://github.com/luxonis/depthai-python.git
cd depthai-python
mkdir build && cd build
cmake ..
cmake --build . --parallel
cd ..
python3 -m pip wheel . -w wheelhouse
pip install wheelhouse/depthai-*

# Test DepthAI with a OAK plugged to your new M1 Mac
cd examples
nano install_requirements.py
    Remove code of block (3 lines) starting with: if thisPlatform == "arm64" and platform.system() == "Darwin":
    Remove code of block (48 lines) starting with: if not args.skip_depthai:
python3 install_requirements.py
python3 ColorCamera/rgb_preview.py

Enjoy!

eimarinb

Hi! Thanks for the code! I'm, new to my M1 mac and macOS and I'm having some problems when I run this part
cmake ..
I get this error
CMake Error: CMAKE_C_COMPILER not set, after EnableLanguage CMake Error: CMAKE_CXX_COMPILER not set, after EnableLanguage
I've tried these solutions from stackoverflow , but they haven't helped

sudo xcode-select --reset
from [1]https://stackoverflow.com/questions/41380900/cmake-error-no-cmake-c-compiler-could-be-found-using-xcode-and-glfw

export SDKROOT=$(xcrun --sdk macosx --show-sdk-path)
xcode-select -p
and selecting the path that I get from there
sudo xcode-select -s /Library/Developer/CommandLineTools
from [2]https://stackoverflow.com/questions/65276102/compilation-errors-with-m1-macbook-pro-and-cmake

mjwhite2529

Thanks for writing the install steps. I ran into the following error while running the cmake ..
The depthai-core/CMakeLists.txt and depthai-core/shared directory didn't get created.
Can you suggest a solution? Thank you.

CMake Error at depthai-core/shared/depthai-shared.cmake:38 (string):
string sub-command SUBSTRING requires four arguments.
Call Stack (most recent call first):
depthai-core/CMakeLists.txt:150 (include)

CMake Error at depthai-core/shared/depthai-shared.cmake:39 (if):
if given arguments:

"STREQUAL" "-"

Unknown arguments specified
Call Stack (most recent call first):
depthai-core/CMakeLists.txt:150 (include)

Here are the contents of the error log:

Compiling the C compiler identification source file "CMakeCCompilerId.c" failed.
Compiler: /Library/Developer/CommandLineTools/usr/bin/cc
Build flags:
Id flags:

The output was:
1
ld: library not found for -lSystem
clang: error: linker command failed with exit code 1 (use -v to see invocation)

Compiling the CXX compiler identification source file "CMakeCXXCompilerId.cpp" failed.
Compiler: /Library/Developer/CommandLineTools/usr/bin/c++
Build flags:
Id flags:

The output was:
1
ld: library not found for -lc++
clang: error: linker command failed with exit code 1 (use -v to see invocation)

weiYao

still no further progress in new M1 chips?I mean native.-_-.

erik

Hello weiYao ,
M1 runners aren't yet supported by Github, that's why we haven't yet added support for native M1, more info here. As soon as that comes out we will also add support for M1 on depthai.
Thank you for understanding, Erik

gregflurry

I am trying to migrate from a working DepthAI/OAK-D on an older (X86-based) MacBook to a new (M1 Max-based) MacBook. Using the standard (and naive, in retrospect) migration approach left me with nothing working, due to several issues, including Python locations and versions, non-functional Homebrew, etc. I think I've got all issues resolved, except for the lack of M1 native DepthAI. Based on this discussion thread and the DepthAI documentation, I seem to have 3 options I can: (1) run the X86 DepthAI using Rosetta; (2) build the M1 native wheels; (3) wait for Luxonis to produce the M1 native wheels.

My needs are currently not urgent; I'm not wild about Rosetta dependencies; from the thread it seems that with my level of system skills I'd never succeed at building M1 native wheels. That leaves waiting for Luxonis to produce the M1 native wheels. Is there an estimate of when that might happen? Thanks!

Brandon

This is not going to be automatically built:
https://github.com/luxonis/depthai-python/pull/604

And sorry for missing your comment @gregflurry . Fortunately this will now be done as soon as this finishes building (i.e. tomorrow).

gregflurry

Brandon This sounds exciting! That said, I don't fully understand. The GitHub link appears to suggest that M1-specific (not universal) wheels will get produced; that I think works for me. I'm unsure what you statement "This is not going to be automatically built" means. Does it refer to some Github restriction?

Assuming the wheels do become available somehow, can you estimate when the relevant documentation will get updated? Thanks.

erik

Hello gregflurry ,
We have just implemented building these wheels today. Could you try it out as well?
python3 -mpip install --extra-index-url https://artifacts.luxonis.com/artifactory/luxonis-python-snapshot-local/ depthai==2.16.0.0.dev0+fc472bd13f9c853148e7a35fc956f4bc0dbf49ae

Thanks, Erik

Brandon

Yes it means that M1 will just install w/out having any other work. Will be equivalent to x86 Mac, or x86 Linux and/or all the standard builds we have now.

So this should be fully out this week when we do another release with this system.