Ultralytics

🌟 Summary
This release enhances the object counting functionality in Ultralytics with configurable counting directions, and improves code efficiency and readability.

📊 Key Changes
- Added a new parameter to set object counting directions ("left_to_right" or "right_to_left").
- Updated counting logic to leverage the new direction attribute.
- Refined object counter visualization and class labeling.
- Minor code clean-ups, including removal of redundant components.

🎯 Purpose & Impact
- 🔄 **Flexibility**: Users can now customize the counting direction, making the feature more versatile across different scenarios.
- 🎯 **Improved Accuracy**: The directionality consideration minimizes miscounts, particularly in bidirectional flows.
- 🧹 **Code Clarity**: Streamlined codebase for better maintainability and readability, benefiting developers when further customizing or understanding the code.

What's Changed
* Update merge-main-into-prs.yml by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/16199
* Distance calculation docs fix by RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/16220
* Update more/missing type qualifiers to lowercase MkDocs admonitions by jk4e in https://github.com/ultralytics/ultralytics/pull/16215
* Add https://youtu.be/Tt_35YnQ9uk to docs by RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/16214
* Update TwoWayTransformer Docs. by JasonG98 in https://github.com/ultralytics/ultralytics/pull/16161
* Fixed failing greetings CI by ambitious-octopus in https://github.com/ultralytics/ultralytics/pull/16223
* `bbox` and `masks` coloring based on `track-ids` by RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/16073
* Non-Deterministic Training Despite deterministic=True by ambitious-octopus in https://github.com/ultralytics/ultralytics/pull/15961
* `ultralytics 8.2.92` vertical `line` counter fix by CharanPrasadK in https://github.com/ultralytics/ultralytics/pull/16163

New Contributors
* JasonG98 made their first contribution in https://github.com/ultralytics/ultralytics/pull/16161

**Full Changelog**: https://github.com/ultralytics/ultralytics/compare/v8.2.91...v8.2.92

🌟 Summary
The v8.2.91 update focuses on renaming the `v10DetectLoss` module to `E2EDetectLoss` for YOLOv10 to address several raised issues.

📊 Key Changes
- 🆕 Renamed the `v10DetectLoss` module to `E2EDetectLoss` in the YOLOv10 model code.
- 🧩 Additional benchmarks introduced for the YOLOv10 model.
- 📊 Updated documentation using macros for clearer and more consistent argument tables.

🎯 Purpose & Impact
- 🔧 **Issue Resolution**: This change directly resolves multiple complaints about module misnaming (as seen in GitHub issues), ensuring smoother integration and functionality for developers using the YOLOv10 model.
- 📈 **Enhanced Testing**: New benchmarks allow for a comprehensive assessment of the YOLOv10's performance, providing end-users detailed insights into model efficiency.
- 📚 **Improved Documentation**: The use of macros streamlines updates and ensures consistency across documentation, making it easier to understand and maintain for both new and experienced users.

This release strengthens the reliability of the YOLOv10 model while offering better support for developers through enhancements in both functionality and documentation clarity.

What's Changed
* Add YOLOv10 to Raspberry Pi CI by lakshanthad in https://github.com/ultralytics/ultralytics/pull/16087
* Update NVIDIA Jetson TensorRT Benchmarks by lakshanthad in https://github.com/ultralytics/ultralytics/pull/16156
* updated the amount of macros available in the macros folder by MatthewNoyce in https://github.com/ultralytics/ultralytics/pull/16086
* `ultralytics 8.2.91` fix `v10DetectLoss` module rename for YOLOv10 by Y-T-G in https://github.com/ultralytics/ultralytics/pull/16148


**Full Changelog**: https://github.com/ultralytics/ultralytics/compare/v8.2.90...v8.2.91

🌟 Summary
The Ultralytics YOLO `v8.2.90` release focuses on improving memory management on macOS devices and updating dependencies to ensure compatibility and stability.

📊 Key Changes
- **Apple MPS Memory Optimization**: Integrated `torch.mps.empty_cache()` for macOS to enhance memory management when using MPS devices.
- **ONNXSlim Dependency Update**: Reverted `onnxslim` to version `0.1.32` to address export issues with YOLOv10 for TFLite.
- **Default Save Behavior**: Adjusted behavior based on the interface: now defaults `save` to `True` for CLI and `False` for Python scripts.

🎯 Purpose & Impact
- **Performance Enhancements**: The memory optimization for macOS users can lead to significant training speed improvements, potentially reducing training time by up to 40%.
- **Stability Improvements**: Updating the `onnxslim` dependency ensures smoother model export processes and prevents compatibility errors with new versions.
- **User Experience**: The change in default saving behavior aims to align with user expectations, automatically saving results in CLI usage, reducing confusion, and providing flexibility in Python environments.

Overall, this release promotes better resource management on macOS, secures compatibility with dependency updates, and fine-tunes the user interface experience for more intuitive interactions. 🚀💡

What's Changed
* Revert to ONNXSlim 0.1.32 by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/16084
* MPS unified memory cache empty by Oil3 in https://github.com/ultralytics/ultralytics/pull/16078
* Fix Visualization Arguments docs table by MatthewNoyce in https://github.com/ultralytics/ultralytics/pull/16061
* `ultralytics 8.2.90` Apple MPS unified memory clearing by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/16085


**Full Changelog**: https://github.com/ultralytics/ultralytics/compare/v8.2.89...v8.2.90

🌟 Summary
The v8.2.89 release of the Ultralytics package brings significant improvements to the object counting capability, enhancing detection accuracy by refining intersection logic and removing redundant parameters.

📊 Key Changes
- Removed unused parameters like `track_thickness`, `line_dist_thresh`, and `cls_txtdisplay_gap`.
- Introduced a new static method `does_intersect` for better intersection checking.
- Improved logic for counting objects crossing a designated line.

🎯 Purpose & Impact
- **Code Simplification**: By eliminating unused parameters, the code is more streamlined and easier to handle.
- **Enhanced Precision**: Object counting is now more accurate due to refined intersection detection methods.
- **User Experience**: Improved reliability in object tracking applications benefits users with more accurate data analysis, making practical deployments more effective.

Additional improvements in the release include updates to documentation, such as the addition of new recognition entries for authors and the inclusion of useful video tutorials to aid in user understanding of specific features like deploying YOLOv8 on NVIDIA Jetson platforms. These enhancements ensure improved transparency, recognition, and usability for both documentation and practical deployment scenarios.

What's Changed
* Add `not.committed.yet` mkdocs author by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/16049
* Add https://youtu.be/wWmXKIteRLA to docs by RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/16063
* Pass `args` when creating validator for classification by Y-T-G in https://github.com/ultralytics/ultralytics/pull/16025
* Fix `_predict_augment` and add warning by Laughing-q in https://github.com/ultralytics/ultralytics/pull/16056
* CoreML export: updated nms flag for end2end models by ambitious-octopus in https://github.com/ultralytics/ultralytics/pull/16058
* Fix gitignore to format Docs datasets by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/16071
* Update to lowercase MkDocs admonitions by MatthewNoyce in https://github.com/ultralytics/ultralytics/pull/15990
* `ultralytics 8.2.89` Increased `line_counter` Solutions accuracy by RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/16033

New Contributors
* MatthewNoyce made their first contribution in https://github.com/ultralytics/ultralytics/pull/15990

**Full Changelog**: https://github.com/ultralytics/ultralytics/compare/v8.2.88...v8.2.89

🌟 Summary
The v8.2.88 release of Ultralytics introduces key refinements, notably simplifying the distance calculation feature and making several enhancements across the codebase, significantly benefitting both developers and end-users.

📊 Key Changes
- **Distance Calculation Overhaul**: The distance calculation in the software now exclusively uses pixel units, eliminating the `pixels_per_meter` metric.
- **Documentation Updates**: Enhanced documentation to align with the modified distance calculation process.
- **Raspberry Pi CI Workflow**: Removed the need for Raspberry Pi CI reboots to prevent resource issues.
- **Dataset Label Fixes**: Corrected several class name typos in the `Objects365.yaml` file.
- **Dependency Upgrades**: Updated the `mkdocs-ultralytics-plugin` dependency to version 0.1.8.

🎯 Purpose & Impact
- **Simplification & Consistency**: By removing the pixel-to-meter conversion, the calculation process is cleaner, reducing user confusion and potential errors 🚀.
- **Stability & Efficiency**: Improved CI stability by eliminating unnecessary Raspberry Pi reboots, leading to more reliable workflows ✨.
- **Accuracy**: Corrected dataset labels ensure data precision, which is crucial for robust training and evaluation outcomes 📝.
- **Documentation Clarity**: Updated guides ensure users can understand and use the new system quickly and efficiently, improving overall user experience 🛠️.
- **Compatibility**: Updating dependencies ensures the integration of the latest features and fixes, enhancing functionality and stability across the system 🔄.

What's Changed
* Update `mkdocs-ultralytics-plugin>=0.1.8` by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/15991
* Remove Raspberry Pi CI reboot by lakshanthad in https://github.com/ultralytics/ultralytics/pull/15987
* Fix 3 `Objects365.yaml` class names by Lornatang in https://github.com/ultralytics/ultralytics/pull/16002
* Add UltralyticsAssistant to mkdocs_github_authors.yaml by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/15992
* Skip `test_workflow` on Windows CI by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/16003
* Update TFLite > LiteRT docs links by lakshanthad in https://github.com/ultralytics/ultralytics/pull/16020
* Add https://youtu.be/w4yHORvDBw0 to docs by RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/16031
* Ultralytics Code Refactor https://ultralytics.com/actions by UltralyticsAssistant in https://github.com/ultralytics/ultralytics/pull/16047
* Update URLs to redirects by UltralyticsAssistant in https://github.com/ultralytics/ultralytics/pull/16048
* Allows any PyTorch install except `torch==2.4.0` on Windows by Burhan-Q in https://github.com/ultralytics/ultralytics/pull/16019
* `ultralytics 8.2.88` Update `distance-calculation` to pixels by RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/15984

New Contributors
* Lornatang made their first contribution in https://github.com/ultralytics/ultralytics/pull/16002

**Full Changelog**: https://github.com/ultralytics/ultralytics/compare/v8.2.87...v8.2.88

🌟 Summary
Version 8.2.87 of Ultralytics introduces significant improvements in queue management, model export workflows, and compatibility with the latest PyTorch, alongside enhancements for hyperparameter tuning processes.

📊 Key Changes
- **Queue Management Improvement:** Adjusted count reset behavior in the video processing algorithm to enhance accuracy.
- **Model Export Update:** Transitioned model export in testing workflow from TorchScript to the more versatile ONNX format.
- **PyTorch Compatibility:** Incorporated support for PyTorch 2.4 by updating the gradient scaler mechanism.
- **Ray Worker Management:** Ensured cleanup of Ray workers post hyperparameter tuning for better resource management.
- **CI Workflow Resilience:** Enabled continued operation on errors in Conda jobs, avoiding workflow interruptions.
- **Slack Notification Update:** Upgraded Slack notification action to enhance messaging capabilities for CI notifications.

🎯 Purpose & Impact
- **Enhanced Accuracy:** By modifying when counts are reset during video frame processing, users can expect more reliable queue management results.
- **Broadened Compatibility:** Moving to ONNX for exports improves interoperability with diverse platforms and accelerates model deployment.
- **Smooth Transition to New PyTorch Versions:** Updating for full compatibility with PyTorch 2.4 ensures users can leverage the latest features and performance enhancements without disruptions.
- **Efficient Resource Use:** Proper shutdown of Ray workers prevents resource leaks and ensures system efficiency, benefiting long-term stability.
- **Uninterrupted CI Workflows:** Allowing CI workflows to continue despite Conda errors prevents minor failures from impacting broader development processes.
- **Improved Communication:** The updated Slack action facilitates better notification management, offering clarity and robustness in developer communications.

These changes not only optimize current functionalities but also prepare the platform for broader adoption and future technological advancements.

What's Changed
* Bump slackapi/slack-github-action from 1.26.0 to 1.27.0 in /.github/workflows by dependabot in https://github.com/ultralytics/ultralytics/pull/15945
* Continue on Conda CI error by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/15958
* Update `test_workflow` to ONNX by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/15974
* Fix `torch.cuda.amp.GradScaler` warning by Laughing-q in https://github.com/ultralytics/ultralytics/pull/15978
* Fix queue `counts` by TechWolf21 in https://github.com/ultralytics/ultralytics/pull/15971
* `ultralytics 8.2.87` Ray `shutdown()` workers after tuning by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/15976

New Contributors
* TechWolf21 made their first contribution in https://github.com/ultralytics/ultralytics/pull/15971

**Full Changelog**: https://github.com/ultralytics/ultralytics/compare/v8.2.86...v8.2.87