PyTorch Hub results.render()

[glenn-jocher]

This PR implements a new .render() method for PyTorch Hub results objects. results.render() will return a list of original images annotated with predicted bounding boxes. This fourth method adds to the existing 3: .show(), .save(), .print(), and can be used with cv2.imshow() for example.

Note that cv2 usage of the images will require a RGB to BGR conversion, i.e.:

results = model(imgs)
im_list = results.render()
im_list = [x[..., ::-1] for x in im_list]  # RGB to BGR
cv2.imshow(im_list[0])

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Enhanced image rendering, improved information display, and boosted thread management in YOLOv5.

📊 Key Changes

• In models/common.py, a render parameter was added to display method, enabling image array updates after drawing.
• A new render method was introduced to return modified images.
• Color-coded hyperparameter logging was introduced in train.py using the colorstr function.
• Logging string formatting was improved for clarity and conciseness in train.py.
• os.environ['NUMEXPR_MAX_THREADS'] was set in utils/general.py to optimize thread usage.
• Minor typo correction in colorstr function from 'undelrine' to 'underline'.

🎯 Purpose & Impact

• 🖼️ Image Rendering: Users can now programmatically modify and utilize images post-processing (e.g., for further analysis or custom overlays).
• 🎨 Visual Logging Enhancement: Improved log readability, which makes debugging and monitoring the training process easier.
• 🚀 Thread Optimization: More efficient CPU usage, potentially improving computational performances on multi-core systems, and avoiding over-subscription of threads.
• ✏️ Logging and Typos: Cleaner and more informative logging, and corrected function spelling offer better user and developer experience.

[Lifeng1129]

Very Good！