Allow PyTorch Hub results to display in notebooks

[glenn-jocher]

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Enhancements for image display checks and error handling in various environments.

📊 Key Changes

• Added warn=True argument to check_imshow calls when initializing webcam in classify/predict.py, detect.py, and segment/predict.py.
• Introduced CHECK_IMSHOW global variable in models/common.py and conditional display methods based on environment compatibility.
• Implemented @TryExcept decorator for image showing regardless of the environment in models/common.py.
• Enhanced error message formatting in utils/__init__.py and utils/autoanchor.py.
• Added is_notebook utility function in utils/general.py to detect Jupyter notebook environments.
• Adjusted check_imshow in utils/general.py to consider more environment types and optionally warn if display is not supported.
• Minor fixes and improvements to readability and error handling.

🎯 Purpose & Impact

• These changes improve the compatibility of image display capabilities across different platforms like Jupyter notebooks, Docker, and SSH terminals.
• Provide clearer, more meaningful error messages and warnings for developers and users regarding display capabilities.
• Improve user experience by adapting the software behavior to various execution environments without interrupting workflow.
• Non-expert users benefit from streamlined functionality, while expert developers obtain better error handling and debugging support.

[vfdev-5]

Shouldn't ipython be an optional dependency ? IMO, it would make sense to remove it for a minimal setup without "displaying" results.

[glenn-jocher]

@vfdev-5 ipython is only used for checking if environment is notebook, clearing notebook cells and showing images in notebook cells. We have in requirements due to the popularity of your Colab notebook.

Does it conflict with anything or is it installing significant other dependencies in your environment?

[vfdev-5]

Thanks for the response, @glenn-jocher

Does it conflict with anything or is it installing significant other dependencies in your environment?

well, in terms of space, ipython + deps does not take much space (~1-2MB), I agree. I start to understand better all the context of this work. For demos in colab etc it totally makes sense. In terms of headless server usage (probably less frequent?), it feels weird to install ipython, matplotlib, seaborn (it is just my opinion).

[glenn-jocher]

@vfdev-5 got it. Much of this is the difference between training support, visualization support, export support etc. We try to strike the right balance but it's impossible to please everyone.