New Ultralytics Release: YOLO v8.3.78 Launches YOLO12!
We are thrilled to announce the release of Ultralytics v8.3.78. This update introduces the groundbreaking YOLO12 model, along with numerous enhancements, bug fixes, and documentation improvements. Here’s what you need to know:
Summary
With v8.3.78, we’ve expanded the YOLO family with YOLO12, combining cutting-edge attention mechanisms with Ultralytics’ innovation. YOLO12 delivers state-of-the-art accuracy, efficiency, and versatility for computer vision tasks like object detection, segmentation, pose estimation, and more.
Key Changes
1. YOLO12 Models
- New Architecture: Featuring Area Attention, R-ELAN, and FlashAttention for enhanced accuracy and efficiency.
- Expanded Task Support: Handles detection, segmentation, classification, pose estimation, and OBB (Oriented Bounding Box) tasks seamlessly.
- Variants: Available in sizes (
n,s,m,l,x) to suit various computational needs—from edge devices to high-performance cloud environments. - Benchmark Performance: Outperforms previous YOLO versions (YOLO10/YOLO11) and competitors in terms of mAP (mean Average Precision) and efficiency.
2. Documentation Updates
- Detailed YOLO12 documentation with benchmarks, guidance, and use-case examples.
- Inclusion of an informative YouTube demo.
3. Bug Fixes & Code Enhancements
- ONNX Runtime Fixes: Resolved device-handling errors for smoother ONNX workflows (#19322).
- Simplified Exports: Streamlined TFLite export logic by removing unused parameters (#19319).
- Refactored Code: Improved readability and maintainability (#19329).
Purpose & Impact
-
Why YOLO12?
- YOLO12 represents a bold leap into the future of computer vision, offering an attention-centric design to achieve superior accuracy without compromising speed or efficiency.
-
What Does This Mean for You?
- A better and faster model selection process for your projects.
- Access to cutting-edge technology for deployment on diverse devices, including edge hardware.
- Enhanced workflows with robust export formats (e.g., TFLite, ONNX).
What’s Changed
- Remove unused parameters from
export_tfliteby @Y-T-G in #19319 - Fix ONNX RuntimeError with dynamic WorldModel by @Y-T-G in #19322
- Add YOLO12 model info by @Laughing-q in #19328
- Add YouTube demo to docs by @RizwanMunawar in #19331
- Refactor with simplifications by @glenn-jocher in #19329
- Update
model_data.pyby @RizwanMunawar in #19330 - YOLO12 models added by @Laughing-q in #19325
For a full list of changes, visit our Changelog or the Release Page.
Try v8.3.78 Today!
Ready to experience YOLO12 in action? Update your Ultralytics package by running:
pip install ultralytics --upgrade
We value your feedback! Let us know your thoughts, share your results, or report any issues in the Discussions section.
Thank you for being a part of the YOLO community. This release is your toolset to push the boundaries of what’s possible in AI vision !