🌟 Summary
This release enhances Sony IMX device support, improves pose estimation export, and makes visual prompting workflows smoother—while also refining documentation and confusion matrix plotting. 🚀🤖
📊 Key Changes
- Sony IMX Export Improvements:
- Added
mct-quantizers>=1.6.0
as a required dependency for IMX model export to ensure compatibility. - IMX export now supports both object detection and pose estimation models (previously only detection).
- Updated export logic and post-processing to handle pose estimation outputs on IMX devices.
- Added
- Documentation Enhancements:
- Expanded Sony IMX500 integration docs with detailed pose estimation examples, clearer setup instructions, and updated folder structure visuals.
- Improved IBM watsonx integration docs for better clarity.
- YOLOE documentation now explains that visual prompt class selections persist for future predictions and exports.
- Confusion Matrix Plotting:
- Fixed class label handling for confusion matrices with 100+ classes, ensuring plots remain readable and internal data stays unchanged.
- Workflow & Dependency Updates:
- Upgraded Slack GitHub Action to v2.1.1 for improved notification reliability.
- Bumped Ultralytics package version to 8.3.167.
🎯 Purpose & Impact
- Broader IMX Device Support:
- Users can now export and run both detection and pose estimation models on Sony IMX hardware, unlocking new edge AI use cases.
- The added dependency ensures smoother, more reliable model exports with fewer compatibility issues.
- Easier Visual Prompting:
- Once classes are set using visual prompts in YOLOE, users don’t need to re-supply them for future predictions or exports—streamlining workflows.
- Better Usability & Accessibility:
- Enhanced documentation and improved confusion matrix plots make it easier for users of all levels to understand, deploy, and interpret their models.
- Improved Reliability:
- Workflow and dependency updates reduce the risk of export or notification failures, making the development experience more robust.
In summary:
This update makes Ultralytics models more versatile on Sony IMX devices, simplifies visual prompting, and improves the overall user experience for both developers and non-experts. 🎉
What's Changed
- Fix
self.names
overwriting inConfusionMatrix
by @Laughing-q in #21356 - YOLOE: Document ability to persist
visual_prompts
when running prediction withrefer_image
by @Y-T-G in #21368 - Bump slackapi/slack-github-action from 2.1.0 to 2.1.1 in /.github/workflows by @dependabot[bot] in #21370
- Support
IMX
export and inference for Pose Estimation by @ambitious-octopus in #20196 ultralytics 8.3.167
Fix Sony IMX exportmct-quantizers>=1.6.0
dependency by @ambitious-octopus in #21404
Full Changelog: v8.3.166...v8.3.167