Open vsivsi opened 1 week ago
👋 Hello @vsivsi, thank you for bringing this up and for your detailed report 🚀! We appreciate your thorough analysis and inclusion of a Minimal Reproducible Example (MRE) 🐛. This will help us investigate the potential NumPy compatibility issue on macOS more efficiently.
To assist further, we'd suggest checking the following while we look into this:
Ensure you are using the latest ultralytics
package and related dependencies, as recent bug fixes or compatibility updates might already address the issue. You can upgrade with:
pip install -U ultralytics
Additionally, confirm that your environment meets the package requirements (Python>=3.8, PyTorch>=1.8).
YOLO can also be run in verified environments where all dependencies are preconfigured:
For heavy workloads, verified cloud environments such as AWS, Google Cloud, or Docker Images might be beneficial.
As noted in your report, downgrading to a compatible version of NumPy (numpy<2
) might temporarily mitigate the issue:
pip install "numpy<2"
This could serve as a workaround while our team investigates the compatibility with NumPy 2.x further.
If you'd like to share this issue with the community or discuss similar experiences, consider engaging with others on:
If this badge is green 💚, all Ultralytics CI tests including MacOS are passing. CI runs daily to ensure cross-platform compatibility with all supported modes and tasks.
An Ultralytics engineer will follow up on this issue soon! Let us know if you observe any additional behavior that might help us reproduce this more effectively. Thanks for helping us improve Ultralytics! 😊
You can try upgrading PyTorch
Search before asking
Ultralytics YOLO Component
Install
Bug
This PR https://github.com/ultralytics/ultralytics/pull/18945 seems to have introduced an Numpy incompatibility within the Ultralytics package on MacOS: e.g.
Environment
Minimal Reproducible Example
Additional
No response
Are you willing to submit a PR?