# New Release: Ultralytics v8.3.39 **URL:** \ **Category:** Discussion\ **Tags:** releases, announcements, ultralytics-official\ **Created:** [November 29, 2024, 1:10pm UTC](https://community.ultralytics.com/t/new-release-ultralytics-v8-3-39/552 "2024-11-29T13:10:09Z")\ **Posts on this page:** 1\ **Page:** 1
**Author:** ![glenn-jocher](https://sea1.discourse-cdn.com/flex001/user_avatar/community.ultralytics.com/glenn-jocher/32/83_2.png) [@glenn-jocher](https://community.ultralytics.com/u/glenn-jocher)\ **Post date:** [November 29, 2024, 1:10pm UTC](https://community.ultralytics.com/t/new-release-ultralytics-v8-3-39/552/1 "2024-11-29T13:10:09Z")
## 🚀 Ultralytics Release v8.3.39: Smarter, Faster, Better Models! We’re thrilled to announce the release of **Ultralytics v8.3.39**! This update brings significant upgrades to enhance functionality, accuracy, and user experience, making it one of our most impactful releases yet. Below is a summary of the key highlights and improvements. * * * ## 🌟 Summary The **v8.3.39** release focuses on refining model behavior, improving tooling, and enhancing the overall usability of the platform. With new features, fixed issues, and optimized workflows, this version offers something for everyone—from developers to data scientists. Don’t miss out—explore what’s new below! 🌟 * * * ## 📊 Key Changes ### 🧠 Fixed Classification Validation Loss - Improved the consistency and accuracy of classification validation. - Enhanced the application of `softmax` operations, limited only to appropriate scenarios, ensuring precise model behavior. ### 🎯 New “Classes” Argument for Training - Added a `classes` parameter for training configuration, empowering users to train models on selected class IDs only! This allows for focused and highly specific model training workflows. ### 🎥 Enhanced Video Annotation Tool: “Sweep Annotation” - Introducing a dynamic utility that enables users to annotate video data by tracking objects along an interactive sweep line. Great for advanced tracking use cases! ### 🖼 Improved Color Handling for LibTorch Inference - Added **BGR to RGB conversion** for the C++ LibTorch inference example, significantly improving color accuracy and ensuring seamless compatibility for YOLO results. ### 📚 Documentation Upgrades - Enhanced clarity and accessibility in documentation, including: - Clickable **YOLO11 performance plots** that link directly to docs. - High-quality **tutorial videos** across the platform. - Improved readability and consistency in `README` files. - Corrected nomenclature by transitioning all references of `YOLOv11` to the proper term, **YOLO11**. ### ⚙ Code Refinements - Refined segmentation handling by improving out-of-bounds coordinate clipping—fewer errors, smoother results. - Simplified attribute access for the Model class using the ` __getattr__ ` method—now you can easily access model attributes like `stride` or `task`. - Improved logging for a more seamless developer experience. * * * ## 🎯 Purpose & Impact This update improves accuracy, usability, and functionality across multiple areas: - **Higher Accuracy** : Fixed classification validation scaling issue to deliver better metrics during validation phases. - **Training Flexibility** : The new `classes` filter lets you pinpoint specific class IDs, optimizing resources and driving focus in model training. - **Streamlined Annotation** : The new sweep annotation tool is a game-changer for video annotation workflows! - **Better Inference Quality** : Proper color handling in LibTorch boosts accuracy for users in C++ environments. - **Improved User Onboarding** : Revamped documentation, clickable visual aids, and tutorial videos make starting with Ultralytics easier than ever! * * * ## 📋 What’s Changed - **Documentation Improvements** : See [#17806](https://github.com/ultralytics/ultralytics/pull/17806), [#17816](https://github.com/ultralytics/ultralytics/pull/17816), [#17863](https://github.com/ultralytics/ultralytics/pull/17863) - **Improved Segmentation Handling** : See [#17810](https://github.com/ultralytics/ultralytics/pull/17810), [#17856](https://github.com/ultralytics/ultralytics/pull/17856), [#17864](https://github.com/ultralytics/ultralytics/pull/17864) - **Enhanced Training Control (`classes`)**: See [#17856](https://github.com/ultralytics/ultralytics/pull/17856) - **Fixed LibTorch Color Handling** : See [#17864](https://github.com/ultralytics/ultralytics/pull/17864) - **Video Annotation Tool** : See [#17742](https://github.com/ultralytics/ultralytics/pull/17742) Full changelog available [here](https://github.com/ultralytics/ultralytics/compare/v8.3.38...v8.3.39). * * * ## 🙌 How to Get Started Try **v8.3.39** now and experience improved tools, better performance, and smoother workflows. Whether you’re training, validating, annotating, or deploying, this release has updates just for you! To upgrade: ```bash pip install ultralytics --upgrade ``` Check the latest tutorials and documentation at [Ultralytics Documentation](https://docs.ultralytics.com). * * * ## ✋ We’d Love Your Feedback! Your feedback helps us improve. Let us know your thoughts and experiences with v8.3.39! Share your journey on [GitHub Discussions](https://github.com/ultralytics/ultralytics/discussions). Big thanks to our amazing developers and contributors who made this release possible. Explore the changes, and as always, visit our [GitHub Repository](https://github.com/ultralytics/ultralytics) to contribute or report issues. Let’s continue to innovate, together. 🚀