Error in utils/segment/general masks2segments()

[paulguerrie]

When running segmentation predict on gpu, the conversion from tensor to numpy fails. Calling .cpu() before calling .numpy() solves this problem.

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Enhanced compatibility for YOLOv5 segmentation mask processing.

📊 Key Changes

• Modified the data handling in segment conversion to ensure tensor operations are CPU compatible.

🎯 Purpose & Impact

• Purpose: The change ensures that the segmentation mask conversion process can be executed on systems without GPU support by explicitly moving the mask tensor to the CPU.
• Impact: This improves the flexibility of the code, allowing it to run on a wider range of hardware, particularly benefiting users who may be working with CPU-only environments. It could potentially increase the accessibility of YOLOv5 for users with less powerful machines. 🖥️🚀

[glenn-jocher]

@paulguerrie thanks for the PR! I'm not able to reproduce any issues with GPU segmentation inference. Can you help us by providing the commands you are using that produce the error for you? Thanks!

[paulguerrie]

Of course! Sorry, should have included steps to reproduce in the PR. Here's what I ran in a google colab and the resulting error:

Also, noticing that other runs of segment/predict.py do not fail in the same colab. (Using the Roboflow tutorial colab: https://colab.research.google.com/drive/1JTz7kpmHsg-5qwVz2d2IH3AaenI1tv0N?usp=sharing)

[glenn-jocher]

@paulguerrie ah yes, I'm able to reproduce now!

PR is merged. Thank you for your contributions to YOLOv5 🚀 and Vision AI ⭐