Announcing Ultralytics v8.3.90 Release!
We’re excited to announce the release of Ultralytics v8.3.90, packed with crucial updates, performance boosts, and enhancements to make your experience with YOLO models even better! This release focuses on improving memory calculations for Apple MPS devices, refining YOLO model architectures, and polishing documentation to ensure smoother workflows for developers and users alike.
Here’s what’s new:
Summary of Improvements
- MPS Memory Usage Fix: Corrected memory tracking for Metal Performance Shaders (MPS) using improved calculations.
- YOLO Model Optimizations: Streamlined YOLO11, YOLOv8, and YOLOv9 by reducing layer complexity for enhanced efficiency.
- Polished Documentation: Improved clarity and consistency in documentation with detailed method and parameter descriptions.
- Enhanced Logging: Better customization and debugging through refined logging functionality.
- Bug Fixes for C++ Inference: Resolved issues with input dimensions in YOLOv8’s C++ example code.
- Default Solution Support: Added fallback to “count” solution in command executions for smoother handling.
New Features and Improvements
1. MPS Memory Calculation Fix
Apple MPS device users now benefit from accurate memory usage tracking with seamless training and inference performance.
- Contribution: Fix MPS
get_memory()
error by @Y-T-G.
2. YOLO Model Efficiency Updates
YOLO11, YOLOv8, and YOLOv9 have been optimized by reducing unnecessary layers. This balances performance and speed, making them faster and lighter.
- Contribution: Fix layer counts in model YAMLs by @Y-T-G.
3. Improved Documentation
We’ve enhanced the documentation with clearer language, better formatting, and consistent parameter descriptions, making adoption and usage easier for everyone.
- Contributions:
4. Logging Enhancements
Added improved control over verbose logging messages, ensuring a better debugging experience.
- Contribution: Fix verbose for Solutions by @RizwanMunawar.
5. C++ Inference Bug Fixes
YOLOv8’s C++ example code has been corrected to handle input image dimensions appropriately.
- Contribution: Fix YOLOv8 C++
formatToSquare
by @matriox1003.
6. Default Solution Handling
Improved execution fallback by defaulting to the “count” solution when no specific name is supplied in YOLO commands.
- Contributions:
Impact & Benefits
- User-Friendly Enhancements: Apple MPS users get more reliable memory tracking, and improved documentation aids in faster onboarding.
- Performance Boost: YOLO architecture updates provide higher efficiency across tasks, including detection, segmentation, and pose estimation.
- Developer Convenience: Bug fixes and solution enhancements simplify workflows for a more seamless setup process.
Detailed Highlights
Full Changelog: Check out the complete list of changes in this update and how they elevate your experience:
Changelog for v8.3.90
Release Link: Explore the release on GitHub for a deeper dive into the updates:
Ultralytics v8.3.90
Try It Out and Share Your Feedback!
We encourage you to try the latest release and explore these updates firsthand. Your feedback is invaluable for guiding future improvements. Feel free to start a discussion or raise issues on our GitHub repository.
Thank you for being an integral part of the YOLO community. We’re excited to see what you achieve with Ultralytics v8.3.90!
— The Ultralytics Team