New Release: Ultralytics v8.3.38

Discussion
,
1

:tada: Announcing Ultralytics v8.3.38 Release!

We’re excited to share the release of Ultralytics v8.3.38, packed with new features and enhancements to make your computer vision journey even smoother and more powerful. This update focuses on improving video segmentation, memory efficiency, and compatibility across platforms. Whether you’re a seasoned veteran or just getting started, there’s something here for everyone. :rocket:

:star2: Summary

The v8.3.38 release introduces the groundbreaking SAM2VideoPredictor for interactive video segmentation and tracking, alongside general optimizations for memory management, streamlined configuration, enhanced device compatibility, and documentation updates. It’s all about making your experience faster, smarter, and more interactive.

:rocket: New Features & Improvements

:film_projector: Enhanced Video Segmentation with SAM2VideoPredictor

  • The SAM2VideoPredictor is a game-changer for video processing. It allows for:
    • Object tracking and segmentation in videos.
    • Interactive prompts for user-defined segment modifications.
  • Achievements:
    • Non-overlapping masks for cleaner segmentations.
    • Improved memory management for faster video inference.
    • Dynamic input adjustments for greater control.

:hammer_and_wrench: General Improvements

  • Device Compatibility: Broader support for NVIDIA Jetson devices.
  • Configuration Simplification: Removal of unused parameters like label_smoothing for more user-friendly workflows.
  • Memory Optimization:
    • Improved resource handling during video processing, boosting efficiency across devices.
  • Documentation Refresh: Enhanced clarity and usability in code and documentation.

:wrench: Notable Fixes

We addressed several aspects to ensure a seamless user experience:

  • Fixed issues with the RepC3 module, improving RT-DETR models. #17086
  • Adjusted device recognition to support additional Jetson platforms. #17770
  • Resolved DLA export inconsistencies. #17765
  • Enhanced frame conversions for SAHI inference. #17795

:mag: Details on Key Contributions

Here are some key pull requests for this release, showcasing the amazing work by our contributors:

:star2: A huge thank you to all the contributors for their incredible efforts!

New Contributors:

  • @Andrewymd: First contribution improved RT-DETR models. :tada:
  • @arun477: Made first-ever contribution to library refinement. :tada:

:memo: Full Changelog & Resources

:person_raising_hand: Try It Now!

Ready to experience the new features? Update to v8.3.38 with:

pip install ultralytics --upgrade

Let us know how it works for you, and don’t hesitate to provide feedback, report issues, or suggest enhancements! Together, we can build better tools for computer vision. :star2:

Happy modeling,
The Ultralytics Team :computer: