Improved detect.py timing

[glenn-jocher]

Updated detect.py timing to match hub and val.py formats. Allows for simpler FPS calculation for students, i.e. #4737

$ python detect.py
YOLOv5 🚀 v5.0-419-gc5360f6 torch 1.9.0 CPU

Fusing layers... 
Model Summary: 224 layers, 7266973 parameters, 0 gradients
image 1/2 /Users/glennjocher/PycharmProjects/yolov5/data/images/bus.jpg: 640x480 4 persons, 1 bus, 1 fire hydrant, Done. (0.240s)
image 2/2 /Users/glennjocher/PycharmProjects/yolov5/data/images/zidane.jpg: 384x640 2 persons, 2 ties, Done. (0.191s)
Speed: 1.1ms pre-process, 215.5ms inference, 1.1ms NMS per image at shape (1, 3, 640, 640)  <------- NEW
Results saved to runs/detect/exp12

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

Significant refactor of timing logic in inference scripts for enhanced transparency and clarity.

📊 Key Changes

• Removed the global import of the time module in detect.py.
• Introduced a local, more granular timing system that tracks three distinct stages.
• Changed timing print statements to reflect only the time spent on the inference process, excluding pre- and post-processing times.
• Standardized and cleaned up timing across both detect.py and val.py for consistency.

🎯 Purpose & Impact

• The purpose of these changes is to provide users with more accurate and detailed insights into the model's inference performance 🚀.
• Users will benefit from a clear understanding of where time is spent during model operation, allowing for better optimization and evaluation 🕒.
• This level of detail is especially important for use cases where inference speed is critical, such as real-time applications 📈.

[glenn-jocher]

Divide 1000ms by the inference speed shown to arrive at a mean FPS for all images/frames run. In the above for example:

Speed: 1.1ms pre-process, 215.5ms inference, 1.1ms NMS per image at shape (1, 3, 640, 640)  <------- NEW

1000ms / 215.5ms = 4.7 FPS

[stccb]

Sorry,Why the fps is 1000/inference_time is not （1000/(inference_time+NMS+prediction time)）?