Announcing Ultralytics Release v8.3.87
We are thrilled to unveil the latest Ultralytics release, v8.3.87, packed with exciting new features, improvements, and fixes to enhance your experience with our models. Read on to discover what’s new and why you should upgrade today!
Summary
The v8.3.87 release introduces a new Results.to_html() method for exporting inference results as HTML, making it easier to share and visualize results in a web-friendly format. This update also includes key usability improvements, expanded compatibility for ARM systems, smarter GPU memory management, and enhanced documentation.
New Features
HTML Export for Results
- A brand new
to_html()method allows you to convert your inference results into HTML format, perfect for sharing and showcasing detection outputs. - Simplifies result visualization in browser-based applications.
Thanks to: @RizwanMunawar
Pull Request
Improvements
Documentation Enhancements
- Added a new dedicated page for the COCO128 dataset explanation to guide developers during testing and debugging.
- Improved clarity and consistency of method descriptions across the docs.
Thanks to: @lakshanthad
COCO128 PR
Compatibility Upgrades
- ARM Support: Enabled Ubuntu ARM GitHub CI runners for better ARM-based system compatibility and performance.
Thanks to: @lakshanthad
ARM PR - OpenVINO Version Constraints: Applied restrictions to prevent version conflicts and enhance stability.
Thanks to: @aleksandr-mokrov
OpenVINO PR - Faster ARM64 Docker Builds: Switched to GitHub-hosted ARM64 Ubuntu runners, reducing build times from 8 minutes to 2 minutes.
Thanks to: @lakshanthad
Docker PR
Optimized Performance
- Memory Management: Optimized GPU memory clearing by triggering only when utilization exceeds 90%, minimizing unnecessary slowdowns.
Thanks to: @Y-T-G
GPU Memory PR
Bug Fixes
- Resolved an Out-of-Bounds Bounding Box issue in MNN examples.
Thanks to: @jules-ai
MNN Example PR - Fixed multi-stream video saving, ensuring no overwriting occurs when handling results.
Thanks to: @XevenQC
Multi-Stream Fix PR
Additional Enhancements
- Comet Integration: Added logging support for segmentation annotations, enabling richer project tracking capabilities for segmentation tasks.
Thanks to: @yaricom
Comet PR - Updated FLOPs Calculations in classification models to align with the default image size of 224 pixels, improving metrics accuracy.
Thanks to: @Y-T-G
FLOPs PR
A Warm Welcome to New Contributors
- @decahedron1 - Excluded
ort.pyke.iofrom link checking.
Contribution PR - @aleksandr-mokrov - Constrained OpenVINO versions for compatibility.
Contribution PR - @XevenQC - Fixed result saving for multi-stream detection.
Contribution PR
Thank you for your contributions and welcome to the YOLO family!
Try It Now!
Upgrade to v8.3.87 and experience the improvements firsthand. We value your feedback to continue enhancing Ultralytics tools for the community. Let us know your thoughts and feel free to open an issue for any suggestions or discoveries.
Full Changelog: Compare Changes
Your enthusiasm and feedback fuel our innovation—thank you for growing with us! 