Ultralytics YOLO v8.3.62 Released! 
We’re excited to announce the release of Ultralytics YOLO v8.3.62, bringing significant updates, fixes, and usability improvements! This version focuses on deterministic data augmentation, better developer experience, and overall reliability enhancements.
Summary
Ultralytics v8.3.62 introduces improvements crucial for predictable training processes, repository maintenance, and stability, ensuring smoother workflows and enhanced user experience. Here’s a quick glance at the release highlights:
- Deterministic data augmentations with
albumentations>=1.4.21. - Enhanced organization with renamed
.ymlworkflow files. - Fixes for occasional dataloader freezes during repeated training runs.
- Updated licensing and documentation headers for a professional and polished product.
Key Changes
New Features
- Non-Deterministic Transforms Resolved
This update addresses randomness in data augmentation pipelines, ensuring reproducible and consistent model training outcomes by introducing support for random seeding.
Workflow and Documentation Enhancements
- GitHub workflow files renamed from
.yamlto.ymlfor better uniformity. - Licensing headers standardized across all source and config files for clarity and consistency.
- Updated year metadata to 2025 in the documentation for relevance and accuracy.
Bug Fixes
- Fixed sporadic freezes occurring in dataloaders during consecutive training instances, ensuring stable and reliable training environments.
Code Clean-Up
- Improved hyperparameter mutation by optimizing access to internal data structures, leading to a more streamlined codebase.
Purpose & Impact
This release focuses on ensuring reliability and productivity for Ultralytics YOLO users:
- Deterministic Training: Gain consistent results for debugging and evaluation.
- Developer-Friendly Updates: Simplify contributions with better organized workflows and standardized headers.
- Improved Stability: Run longer training sessions without issues.
- Polished Representation: Showcases Ultralytics as a forward-thinking AI platform with modernized documentation and licensing practices.
Whether you’re a developer, researcher, or student, these changes make your work with YOLO faster, smoother, and frustration-free!
What’s Changed
Below is a list of merged PRs and the contributors behind them:
- Consistent workflow file suffix
.ymlby @glenn-jocher in #18668 ci.ymlanddocker.ymlrename by @glenn-jocher in #18671- Fix MNN example missing BGR-to-RGB conversion by @jules-ai in #18689
- Remove unused
dictvalues in.items()with.values()by @Kayzwer in #18651 - Updated year metadata to 2025 by @glenn-jocher in #18695
- Standardized license headers in Python files by @pderrenger in #18696
- Fix for sporadic dataloader freezes by @Y-T-G in #18697
- Enhanced headers and comments in TOML/YAML files by @pderrenger in #18698
- Fix non-deterministic transforms using
albumentationsby @Y-T-G in #18701
Welcome New Contributors
A warm welcome and big thanks to @jules-ai for making their first contribution to the repository!
How to Get Started
Update to the latest release by running:
pip install ultralytics --upgrade
Explore our Release Notes for more details.
We’d love to hear your thoughts! If you have any feedback or run into issues, drop by the Discussions or Issues section.
Happy coding and thank you for being part of the YOLO community!