ultralytics 8.1.43 40% faster ultralytics imports

[glenn-jocher]

🛠️ PR Summary

_{Made with ❤️ by Ultralytics Actions}

🌟 Summary

This PR introduces various updates across testing, data handling, and utility modules in the Ultralytics repository.

📊 Key Changes

• Testing Enhancements: Simplified the tensor inference in tests by replacing complex OpenCV and torchvision transformations with direct PyTorch tensor creation.
• Module Refactor: Moved the ClassificationDataset class definition for clarity and improved import efficiency.
• Version Bump: Updated the package version to 8.1.43.
• Dependency Management: Reduced direct imports from torchvision and other heavy libraries within function scopes to accelerate startup times.
• Code Streamlining: Removed unused imports and updated code for better readability and performance. For example, in test and data augmentation functions, less verbose and more efficient data handling approaches are now used.

🎯 Purpose & Impact

• Enhanced Test Reliability: The updates to testing scripts ensure more straightforward and reliable testing processes.
• Improved Import Efficiency: By deferring heavy imports, the updates contribute to faster load times for the Ultralytics package, enhancing the user experience, especially in environments where startup time is critical.
• Future-Proofing: Cleaning up and streamlining the codebase helps maintain the package more efficiently and lays a solid foundation for future enhancements.
• User and Developer Experience: These changes benefit both users of the Ultralytics package by providing a smoother experience and developers by easing the process of understanding and contributing to the codebase.

In summary, this PR brings forth important updates that streamline operations, enhance performance, and improve the overall user and developer experience of the Ultralytics repository. 🚀

[codecov]

Codecov Report

Attention: Patch coverage is 31.81818% with 75 lines in your changes are missing coverage. Please review.

Project coverage is 37.96%. Comparing base (99c61d6) to head (9bf4422).

❗ Current head 9bf4422 differs from pull request most recent head 6b83a36. Consider uploading reports for the commit 6b83a36 to get more accurate results

Files	Patch %	Lines
ultralytics/data/dataset.py	10.60%	59 Missing ⚠️
ultralytics/data/augment.py	20.00%	4 Missing ⚠️
ultralytics/data/explorer/gui/dash.py	0.00%	2 Missing ⚠️
ultralytics/models/fastsam/prompt.py	0.00%	2 Missing ⚠️
ultralytics/utils/checks.py	50.00%	2 Missing ⚠️
ultralytics/data/explorer/explorer.py	50.00%	1 Missing ⚠️
ultralytics/data/explorer/utils.py	0.00%	1 Missing ⚠️
ultralytics/models/sam/predict.py	50.00%	1 Missing ⚠️
ultralytics/models/yolo/classify/train.py	0.00%	1 Missing ⚠️
ultralytics/utils/benchmarks.py	0.00%	1 Missing ⚠️
... and 1 more

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #9547      +/-   ##
==========================================
- Coverage   37.99%   37.96%   -0.03%     
==========================================
  Files         121      121              
  Lines       15277    15281       +4     
==========================================
- Hits         5804     5802       -2     
- Misses       9473     9479       +6     

Flag	Coverage Δ
GPU	37.96% <31.81%> (-0.03%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[ddeevvaa]

how faster it can be?any test?

[glenn-jocher]

@ddeevvaa I still haven't removed it completely, there's one file left, but seems to improve speed from about 1.4s -> 1.0s import

[glenn-jocher]

@ddeevvaa final results are here averaged over 30 imports. Speedup of 25.3% acheived, reduced import time from 1411ms by 357ms to 1054ms on my Macbook Air M2.

EDIT: 542 ms removed from import (1411 ms -> 869 ms 38% faster) after also scoping pandas that had accidentally be de-scoped. This is HUGE!!!!

this PR

• 1.054 s for import ultralytics
• EDIT: 0.869 s after also scoping pandas that had been inadvertently added as top-level Explorer imports

main branch

• 1.411s for import ultralytics

[ddeevvaa]

very strong~
a remarkable improvement,it seems mergence will be soon

[glenn-jocher]

@ddeevvaa thank you so much for the encouraging feedback! 🌟 We're thrilled by the performance improvements as well and can't wait for the merge to happen. It's all about making YOLOv8 better and faster for everyone. Keep an eye out for updates soon!