[Beta Support]: Working 0.17 Mac Silicon Installation AND Yolov9 #21290

mjalindahl · 2025-12-15T08:16:49Z

mjalindahl
Dec 15, 2025

Describe the problem you are having

Start by using the "ghcr.io/blakeblackshear/frigate:0.17.0-beta1-standard-arm64" image (the beta docs say to use the stable version; that won't work for 0.17)
Download the Apple Silicon Detector (https://github.com/frigate-nvr/apple-silicon-detector)
The documentation says to right-click and click "Open" to bypass the MacOS security checks. That hasn't been a thing for a while. Open Terminal, cd into your Downloads directory (most likely) and run ~/Downloads/FrigateDetector\ 2.app/Contents/MacOS/FrigateDetector
This will actually run the Frigate Detector, and you should see an output that says:
2025-12-14 08:57:25,188 - __main__ - INFO - Cleanup completed 2025-12-14 08:57:25,189 - __main__ - INFO - ZMQ ONNX client initialized with endpoint: tcp://*:5555 2025-12-14 08:57:25,189 - __main__ - INFO - ZMQ ONNX client started in AUTO mode - waiting for model requests 2025-12-14 08:57:25,189 - __main__ - INFO - Attempting to bind to endpoint: tcp://*:5555 2025-12-14 08:57:25,189 - __main__ - INFO - ZMQ server successfully bound to tcp://*:5555 2025-12-14 08:57:25,189 - __main__ - INFO - Detector is ready to accept model requests and inference requests
It's not made super clear in the documentation, although it is technically there, no models come with this install. You'll need to provide your own. To use Yolov9, run this:

docker build . --build-arg MODEL_SIZE=t --build-arg IMG_SIZE=320 --output . -f- <<'EOF' FROM python:3.11 AS build RUN apt-get update && apt-get install --no-install-recommends -y libgl1 && rm -rf /var/lib/apt/lists/* COPY --from=ghcr.io/astral-sh/uv:0.8.0 /uv /bin/ WORKDIR /yolov9 ADD https://github.com/WongKinYiu/yolov9.git . RUN uv pip install --system -r requirements.txt RUN uv pip install --system onnx==1.18.0 onnxruntime onnx-simplifier>=0.4.1 onnxscript ARG MODEL_SIZE ARG IMG_SIZE ADD https://github.com/WongKinYiu/yolov9/releases/download/v0.1/yolov9-${MODEL_SIZE}-converted.pt yolov9-${MODEL_SIZE}.pt RUN sed -i "s/ckpt = torch.load(attempt_download(w), map_location='cpu')/ckpt = torch.load(attempt_download(w), map_location='cpu', weights_only=False)/g" models/experimental.py RUN python3 export.py --weights ./yolov9-${MODEL_SIZE}.pt --imgsz ${IMG_SIZE} --simplify --include onnx FROM scratch ARG MODEL_SIZE ARG IMG_SIZE COPY --from=build /yolov9/yolov9-${MODEL_SIZE}.onnx /yolov9-${MODEL_SIZE}-${IMG_SIZE}.onnx EOF

You'll need to copy that output - there are two files, one for Yolov9 320x320, and one for Yolov9 640x640. Decide which one you want to use, and copy/rename that file to "/config/model_cache/yolo.onnx" (important) on your Mac - in your main Frigate directory, where you have your config.yaml.
Profit.

I've been running it for ~12 hours now, and it's been absolutely flawless. With 2x 4K cameras and 2x 1080p cameras, running the 320x320 version, I'm averaging ~11ms inference times.

Beta Version

0.17.0-beta1

Issue Category

Installation / Updates

Frigate config file

detectors:
  apple-silicon:
    type: zmq
    endpoint: tcp://host.docker.internal:5555

model:
  model_type: yolo-generic
  width: 320 # <--- should match the imgsize set during model export
  height: 320 # <--- should match the imgsize set during model export
  input_tensor: nchw
  input_dtype: float
  path: /config/model_cache/yolo.onnx
  labelmap_path: /labelmap/coco-80.txt

Relevant Frigate log output

N/A

Relevant go2rtc log output (if applicable)

No response

Install method

Docker CLI

docker-compose file or Docker CLI command

docker run -it -p 5000:5000 -p 8971:8971 --restart=unless-stopped --name frigate -v /Users/[name]/docker/frigate-mac/config:/config -v /Volumes/FrigateMedia:/media/frigate ghcr.io/blakeblackshear/frigate:0.17.0-beta1-standard-arm64

Operating system

Other

CPU / GPU / Hardware

Mac M4 Mini Silicon

Screenshots

No response

Steps to reproduce

No response

Any other information that may be helpful

No response

NickM-27 · 2025-12-15T12:46:18Z

NickM-27
Dec 15, 2025
Collaborator Sponsor

Thanks, I'll update some of the documentation

0 replies

ronbruins · 2025-12-15T22:05:45Z

ronbruins
Dec 15, 2025

I get the below error message when running docker build.

[build 7/10] RUN uv pip install --system onnx==1.18.0 onnxruntime onnx-simplifier>=0.4.1 onnxscript:
0.101 Using Python 3.11.14 environment at: /usr/local
0.440 Resolved 19 packages in 336ms
0.538 Downloading onnx (16.6MiB)
0.538 Downloading ml-dtypes (4.8MiB)
0.538 Downloading onnxruntime (14.5MiB)
3.404 Downloading ml-dtypes
5.285 Downloading onnxruntime
5.498 Downloading onnx
5.665 Building onnx-simplifier==0.4.36
6.173 × Failed to build onnx-simplifier==0.4.36
6.173 ├─▶ The build backend returned an error
6.173 ╰─▶ Call to setuptools.build_meta:__legacy__.build_wheel failed (exit
6.173 status: 1)
6.173
6.173 [stderr]
6.173 :28: DeprecationWarning: Use shutil.which instead of
6.173 find_executable
6.173 fatal: invalid gitfile format: /root/.cache/uv/sdists-v9/.git
6.173 fatal: invalid gitfile format: /root/.cache/uv/sdists-v9/.git
6.173 Traceback (most recent call last):
6.173 File "", line 14, in
6.173 File
6.173 "/root/.cache/uv/builds-v0/.tmpYSvMaw/lib/python3.11/site-packages/setuptools/build_meta.py",
6.173 line 331, in get_requires_for_build_wheel
6.173 return self._get_build_requires(config_settings, requirements=[])
6.173 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
6.173 File
6.173 "/root/.cache/uv/builds-v0/.tmpYSvMaw/lib/python3.11/site-packages/setuptools/build_meta.py",
6.173 line 301, in _get_build_requires
6.173 self.run_setup()
6.173 File
6.173 "/root/.cache/uv/builds-v0/.tmpYSvMaw/lib/python3.11/site-packages/setuptools/build_meta.py",
6.173 line 512, in run_setup
6.173 super().run_setup(setup_script=setup_script)
6.173 File
6.173 "/root/.cache/uv/builds-v0/.tmpYSvMaw/lib/python3.11/site-packages/setuptools/build_meta.py",
6.173 line 317, in run_setup
6.173 exec(code, locals())
6.173 File "", line 67, in
6.173 AssertionError: Could not find "cmake" executable!
6.173
6.173 hint: This usually indicates a problem with the package or the build
6.173 environment.
Dockerfile:7
5 | ADD https://github.com/WongKinYiu/yolov9.git .
6 | RUN uv pip install --system -r requirements.txt
7 | >>> RUN uv pip install --system onnx==1.18.0 onnxruntime onnx-simplifier>=0.4.1 onnxscript
8 | ARG MODEL_SIZE
9 | ARG IMG_SIZE

ERROR: failed to build: failed to solve: process "/bin/sh -c uv pip install --system onnx==1.18.0 onnxruntime onnx-simplifier>=0.4.1 onnxscript" did not complete successfully: exit code: 1

1 reply

NickM-27 Dec 15, 2025
Collaborator Sponsor

add a --platform linux/amd64 to the command

ronbruins · 2025-12-16T00:58:51Z

ronbruins
Dec 16, 2025

Maybe a stupid question, but even on macOS for Apple silicon? On 16 Dec 2025, at 02:22, Nicolas Mowen ***@***.***> wrote: add a --platform linux/amd64 to the command —Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you commented.Message ID: ***@***.***>

2 replies

NickM-27 Dec 16, 2025
Collaborator Sponsor

Yes, because some packages are only pre built for amd64

ronbruins Dec 16, 2025

Thanks! That did the trick!

Landonthi · 2025-12-16T17:48:18Z

Landonthi
Dec 16, 2025

Ran into an issue last night installing the Apple Silicon Detector. Installed latest python version, app started installing then stopped with the following error. Screen shot attached.

6 replies

jctull Dec 18, 2025

It is unclear how you can define the working python version with the binary installer. Any hints on how to force it to use an older version?

NickM-27 Dec 18, 2025
Collaborator Sponsor

If you install python 3.11 using homebrew it will use that

jctull Dec 18, 2025

I ran a python venv @3.11 and used the detector source code to successfully run that bit. Trying to troubleshoot the docker build command currently, not as successfully, but I have podman on my server, which may not be compatible with the command. It seems to be hanging on “[]> []” after I run it. (Edited, said docker-compose, meant podman)

jctull Dec 19, 2025

.

jctull Dec 20, 2025

I installed docker and docker desktop with homebrew, then was able to build the model. I am running into issues with frigate now, but this is a good start. Thanks for sharing your setup.

ngTamAn111103 · 2025-12-20T16:21:58Z

ngTamAn111103
Dec 20, 2025

I've successfully launched it, thank you. Sharing my configuration so others can have a basis: I'm using a Mac mini M4 2024 10CPU 10GPU 24GB 256GB (in the volumes of docker-compose.yml, I point to an external HDD connected via USB-A). I have two cameras with three streams. In detect mode, I set the detection to 1920x1080 and 800x448 for two streams. The "Detector Inference Speed" is around 20.77ms. I'm using yolov9-t-320.onnx and it's been quite good so far. But I will upgrade both the streams and the model later. Thanks again to the person who shared.

7 replies

fittyCent Jan 6, 2026

Ah interesting. How would it know what regions to crop and scale? By motion?

Also, are you downscaling your camera feeds from say a 4k camera or are you using substream? I was downscaling but it was very unstable if i ran it through videotoolbox (mac's hardware) so I'm just using substream now which is lower res than 1280x720

NickM-27 Jan 6, 2026
Collaborator Sponsor

It is complex logic that uses a mix of motion, existing objects, and other data to calculate the regions.

And I personally use sub streams but I use the recommended cameras that provide high quality sub streams.

fittyCent Jan 6, 2026

Ah got it. Dude, I wanna thank you so much for creating the native apple silicon detector. Long story short, i was really hoping that frigate supported hardware detection on apple silicon since the chip is so efficient. I gave up trying, got a linux nuc and it was running at 50W with yolo8 then I read about your work. I just got it set up and it's doing 10W with 3x 4K, 2x 2K, 2x 1080, 1x 8K (all substreams) on an m4 mini. Ridiculously good. Thank you thank you.

NickM-27 Jan 6, 2026
Collaborator Sponsor

Glad to hear that it is working well

NickM-27 Jan 6, 2026
Collaborator Sponsor

And thank you for the coffee :)

Eun · 2025-12-22T14:10:13Z

Eun
Dec 22, 2025

The all in one docker command wasn't working for me so here is the Dockerfile:

FROM python:3.11 AS build
RUN apt-get update && \
    apt-get install --no-install-recommends -y libgl1 && \
    rm -rf /var/lib/apt/lists/*
COPY --from=ghcr.io/astral-sh/uv:0.8.0 /uv /bin/
WORKDIR /yolov9
ADD https://github.com/WongKinYiu/yolov9.git .
RUN uv pip install --system -r requirements.txt
RUN uv pip install --system onnx==1.18.0 onnxruntime onnx-simplifier>=0.4.1 onnxscript
ARG MODEL_SIZE
ARG IMG_SIZE
ADD https://github.com/WongKinYiu/yolov9/releases/download/v0.1/yolov9-${MODEL_SIZE}-converted.pt yolov9-${MODEL_SIZE}.pt
RUN sed -i "s/ckpt = torch.load(attempt_download(w), map_location='cpu')/ckpt = torch.load(attempt_download(w), map_location='cpu', weights_only=False)/g" models/experimental.py
RUN python3 export.py --weights ./yolov9-${MODEL_SIZE}.pt --imgsz ${IMG_SIZE} --simplify --include onnx

FROM scratch
ARG MODEL_SIZE
ARG IMG_SIZE
COPY --from=build /yolov9/yolov9-${MODEL_SIZE}.onnx /yolov9-${MODEL_SIZE}-${IMG_SIZE}.onnx

build with

docker build . --platform linux/amd64 --build-arg MODEL_SIZE=t --build-arg IMG_SIZE=320 --output . -f Dockerfile

0 replies

Uh oh!

[Beta Support]: Working 0.17 Mac Silicon Installation AND Yolov9 #21290

Uh oh!

Uh oh!

Describe the problem you are having

Beta Version

Issue Category

Frigate config file

Relevant Frigate log output

Relevant go2rtc log output (if applicable)

Install method

docker-compose file or Docker CLI command

Operating system

CPU / GPU / Hardware

Screenshots

Steps to reproduce

Any other information that may be helpful

Replies: 6 comments · 16 replies

Uh oh!

NickM-27 Dec 15, 2025 Collaborator Sponsor

Uh oh!

Uh oh!

NickM-27 Dec 15, 2025 Collaborator Sponsor

Uh oh!

Uh oh!

NickM-27 Dec 16, 2025 Collaborator Sponsor

Uh oh!

Uh oh!

Uh oh!

Uh oh!

NickM-27 Dec 18, 2025 Collaborator Sponsor

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

NickM-27 Jan 6, 2026 Collaborator Sponsor

Uh oh!

Uh oh!

NickM-27 Jan 6, 2026 Collaborator Sponsor

Uh oh!

NickM-27 Jan 6, 2026 Collaborator Sponsor

Uh oh!

Replies: 6 comments 16 replies

NickM-27
Dec 15, 2025
Collaborator Sponsor

NickM-27 Dec 15, 2025
Collaborator Sponsor

NickM-27 Dec 16, 2025
Collaborator Sponsor

NickM-27 Dec 18, 2025
Collaborator Sponsor

NickM-27 Jan 6, 2026
Collaborator Sponsor

NickM-27 Jan 6, 2026
Collaborator Sponsor

NickM-27 Jan 6, 2026
Collaborator Sponsor