PoseLandmarker throws error on GPU-less environments (e.g., M1 MacBook Air)

[arlo-engineer]

Have I written custom code (as opposed to using a stock example script provided in MediaPipe)

Yes

OS Platform and Distribution

macOS Sequoia 15.4.1 on M1 MacBook Air (no discrete GPU)

Mobile device if the issue happens on mobile device

No response

Browser and version if the issue happens on browser

Google Chrome 136.0.7103.114

Programming Language and version

TypeScript (with Next.js 15.3.1)

MediaPipe version

0.10.0

Bazel version

No response

Solution

Pose

Android Studio, NDK, SDK versions (if issue is related to building in Android environment)

No response

Xcode & Tulsi version (if issue is related to building for iOS)

No response

Describe the actual behavior

When trying to initialize PoseLandmarker in a Next.js project, I encounter the following error on an M1 MacBook Air (which does not have a discrete GPU), regardless of whether I set delegate to "CPU" or "GPU":

Failed to initialize pose landmarker: Error: StartGraph failed: $Service "kGpuService", required by node mediapipe_tasks_vision_pose_landmarker_poselandmarkergraph__mediapipe_tasks_vision_pose_detector_posedetectorgraph__mediapipe_tasks_components_processors_imagepreprocessinggraph__ImageCloneCalculator, was not provided and cannot be created: emscripten_webgl_create_context() returned error 0; StartRun failed

Describe the expected behaviour

MediaPipe Pose should work correctly on devices without a GPU (e.g., M1 MacBook Air), regardless of the backend used.

Standalone code/steps you may have used to try to get what you need

You can reproduce the issue by cloning the following minimal repository:
👉 https://github.com/arlo-engineer/mediapipe-pose-study

Setup instructions

1. Clone the repository

   git clone https://github.com/arlo-engineer/mediapipe-pose-study.git
   cd mediapipe-pose-study

2. Install dependenciews

   npm install

3. Run the development server

   npm run dev

4. Open your browser and navigate to:
   http://localhost:3000/pose

Environment

• Node.js: v23.9.0
• npm: v10.9.2

Other info / Complete Logs

• The error appears to be WebGL-related.
• It happens both when setting delegate: "CPU" and delegate: "GPU".
• Not sure if this is a bug or an unsupported environment.
• I couldn’t find documentation that explicitly states whether GPU-less environments are unsupported.
• Confirmed that the PoseLandmarker works correctly on a GPU-equipped M4 MacBook Pro environment.

[kuaashish]

Hi @arlo-engineer,

Apologies for the delay, I was occupied with an important things. Could you please confirm if this issue is still ongoing, or if you require assistance with it?

Thank you!!

[arlo-engineer]

Hi @kuaashish ,
The issue is still ongoing, and I would appreciate your continued assistance.

[tyrmullen]

I can confirm that GPU support is assumed for MediaPipe JavaScript vision tasks. Setting delegate to "CPU" will cause the expensive machine learning to be run on the CPU, but some of the auxiliary image pre- and post-processing may still assume GPU support, as is the case for PoseLandmarker, as you've descibed above.

That being said, there is no expectation of discrete versus integrated GPU, and an M1 chip should be more than sufficient for any auxiliary pre- and post-processing we perform, so I would expect that everything should still work for you.

To help narrow things down, perhaps try a very simple PoseLandmarker test on your device using Chrome, so we can see if the issue is Node.js-specific. After that, we can try verifying browser/Node.js WebGL-related settings and can also try manually creating WebGL(2) contexts to try to determine which requested context settings are triggering this issue (since from your comments above and in the repo, it sounds like the issue is probably that the WebGL(2) context is failing to instantiate).