Ultralytics

🌟 Summary
This release, version 8.3.35, introduces enhanced support for models with dynamic shapes in image processing, making model handling more adaptable and efficient. 🚀

📊 Key Changes
- **Dynamic Models Support**: Improved the `pre_transform` function to enable automatic letterboxing when working with models that support dynamic input shapes.
- **Updated Docker Configuration**: Switched Docker's base image to Python 3.11.10 for better consistency and added PaddlePaddle installation for broader compatibility.
- **Documentation Enhancements**: Improved Ray Tune documentation, benchmarking tools, and documentation site usability with a scalable search bar.
- **Cosmetic and Code Maintenance**: Various JavaScript updates for cleaner code structure and updated styles for improved user interaction.

🎯 Purpose & Impact
- **Enhanced Model Handling**: By supporting dynamic shapes, the update ensures that users working with such models benefit from accurate image preprocessing and potentially improved performance.
- **Consistency and Compatibility**: Docker updates aid in consistent environment setup and extend functionality by supporting PaddlePaddle installations.
- **Improved User Experience**: Revised documentation and a smoother search experience make it easier for users to find information and ensure a seamless interaction with the site.
- **Developer-Focused Improvements**: Code and workflow updates facilitate easier maintenance and readability, enabling developers to work more efficiently.

What's Changed
* Add RTDETRv2 in `benchmarks.md` chart 📈 by RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/17635
* Update extra.js by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/17665
* Update Dockerfile-cpu to `ubuntu:latest` by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/17670
* Docs Search Bar improvements by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/17669
* Add missing import to Raytune docs by Y-T-G in https://github.com/ultralytics/ultralytics/pull/17683
* `ultralytics 8.3.35` enable `auto` letterbox if model is `dynamic` by Laughing-q in https://github.com/ultralytics/ultralytics/pull/17687


**Full Changelog**: https://github.com/ultralytics/ultralytics/compare/v8.3.34...v8.3.35

🌟 Summary
The update to version 8.3.34 focuses on improving prediction reliability in the FastSAM model and enhances various internal systems to optimize workflows and accuracy. 🚀

📊 Key Changes
- 🛠️ Enhanced FastSAM model's `prompt` method to handle cases with empty predictions effectively.
- 🔧 Updated GitHub Actions to use `uv` for dependency installation, reducing potential Python packaging issues.
- 📋 Improved project name handling in training setups to fix issues with special characters, ensuring compatibility with systems like W&B.
- 🔄 Revised `v8_transforms` function with better hyperparameter handling using `Namespace`.
- 🚀 Enhanced dataset configuration for RT-DETR with new parameters like `fraction`, `single_cls`, and `classes` to better align with YOLO dataset management.
- 📈 Refined object counting method in heatmaps to use centroids instead of bounding boxes for improved accuracy.

🎯 Purpose & Impact
- ✅ **Reliable Predictions**: The FastSAM model update helps avoid errors during inference when some results are empty, making the prediction process more robust.
- 💡 **Streamlined Workflows**: Switching to `uv` in GitHub Actions enhances dependency management and ensures smoother continuous integration.
- 🗄️ **Project Naming Flexibility**: By reformatting project names, users will face fewer naming issues, particularly when integrating with various external systems.
- 📊 **Improved Handling of Hyperparameters**: Developers benefit from more manageable code and potentially fewer bugs with the new `Namespace` implementation.
- 🎯 **Enhanced Customization**: The dataset improvements allow users more control over the training process, focusing on specific classes and data subsets for faster experiments.
- 👁️‍🗨️ **Better Object Tracking**: The refined object counting mechanism boosts the precision of tracking, enhancing analytics accuracy which can significantly improve object detection applications.

What's Changed
* Update Actions with uv installs by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/17620
* Fix v8_transforms docstring example by Y-T-G in https://github.com/ultralytics/ultralytics/pull/17630
* Fix W&B project name separator compatibility by ArcPen in https://github.com/ultralytics/ultralytics/pull/17627
* Update Slack usage to v2 by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/17631
* Add `fraction`, `single_cls` and `classes` to `RTDETRDataset` by Y-T-G in https://github.com/ultralytics/ultralytics/pull/17633
* Heatmaps bug fix by RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/17634
* `ultralytics 8.3.34` FastSAM non-detection fix by petercham in https://github.com/ultralytics/ultralytics/pull/17628

New Contributors
* ArcPen made their first contribution in https://github.com/ultralytics/ultralytics/pull/17627
* petercham made their first contribution in https://github.com/ultralytics/ultralytics/pull/17628

**Full Changelog**: https://github.com/ultralytics/ultralytics/compare/v8.3.33...v8.3.34

🌟 Summary
The latest release, v8.3.33, primarily focuses on refining object counting in the Ultralytics YOLO framework, boosting accuracy for tracking objects across specified regions.

📊 Key Changes
- **Object Counting Enhancement**: Overhauled the object counting logic by focusing on centroids for more precise tracking, especially in complex shapes and motions.
- **Updated Documentation**: Clarified the `retina_masks` and `device` arguments in the documentation for better user comprehension.
- **Expanded Hardware Compatibility**: Enabled MNN export on Raspberry Pi and NVIDIA Jetson platforms.
- **CI/CD Improvements**: Upgraded GitHub workflow actions for better integration with Codecov and Slack.

🎯 Purpose & Impact
- **Improved Counting Accuracy**: By utilizing centroids over bounding boxes, the update ensures more reliable object tracking and counting, crucial for applications needing high precision. 🎯
- **User Clarity**: Enhanced documentation provides clearer guidelines, helping both novice and expert users understand configuration impacts better.
- **Broader Device Support**: Allowing MNN exports on more devices fosters flexibility and innovation, broadening the community's ability to deploy models on diverse hardware setups.
- **Streamlined Workflows**: Upgrades to GitHub actions contribute to more efficient development cycles and error handling, ensuring smoother operations and faster updates.

What's Changed
* Update `retina_masks` description by Y-T-G in https://github.com/ultralytics/ultralytics/pull/17587
* Enable MNN on RPi and Jetson by lakshanthad in https://github.com/ultralytics/ultralytics/pull/17583
* Bump codecov/codecov-action from 4 to 5 in /.github/workflows by dependabot[bot] in https://github.com/ultralytics/ultralytics/pull/17597
* Bump slackapi/slack-github-action from 1.27.0 to 2.0.0 in /.github/workflows by dependabot[bot] in https://github.com/ultralytics/ultralytics/pull/17596
* Update `device` argument description for benchmark by Y-T-G in https://github.com/ultralytics/ultralytics/pull/17550
* `ultralytics 8.3.33` Solutions counter direction fix by RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/17607


**Full Changelog**: https://github.com/ultralytics/ultralytics/compare/v8.3.32...v8.3.33

🌟 Summary
The release of `v8.3.32` introduces a major new dataset called "Dog-pose", designed for pose estimation tasks, along with some important improvements and fixes.

📊 Key Changes
- **Dog-pose Dataset**: Added a new dataset consisting of approximately 6,000 images with detailed annotations for 24 keypoints per dog, specifically for pose estimation using YOLO11.
- **Documentation Update**: Enhanced guides and introductory materials for the Dog-pose dataset, including usage through Python and CLI examples.
- **Link Fix**: Corrected a broken URL in the Jetson device setup documentation.
- **Workflow Update**: Extended retry delay for link checks in the GitHub workflow to enhance reliability.
- **Efficiency Fix**: Improved conditional logging for WandB reporting by checking the availability of plot data.

🎯 Purpose & Impact
- 🐕 **Enhanced Pose Estimation**: The Dog-pose dataset greatly expands capabilities in animal pose estimation, useful in fields like veterinary research and animal behavior analysis.
- 📘 **User-Guidance**: Updated documentation makes it easier for users to leverage the new dataset effectively in their projects.
- 🔧 **Improved Accessibility**: Fixing documentation links enhances user experience by providing direct access to the correct setup resources.
- 🕒 **Optimized Workflow**: Longer delays between retries in automated link checks reduce server loads and improve the reliability of workflows.
- 🎨 **Efficient Resource Use**: The logging enhancement prevents the saving of unnecessary plots, optimizing storage and improving artifact management in model training.

What's Changed
* Fix broken Jetson Doc URL by lakshanthad in https://github.com/ultralytics/ultralytics/pull/17549
* Update links.yml to 900s delay by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/17576
* Fix: wandb reporting error if no positive examples by Jamil in https://github.com/ultralytics/ultralytics/pull/17544
* `ultralytics 8.3.32` New Dog-Pose dataset by RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/17556

New Contributors
* Jamil made their first contribution in https://github.com/ultralytics/ultralytics/pull/17544

**Full Changelog**: https://github.com/ultralytics/ultralytics/compare/v8.3.31...v8.3.32

🌟 Summary
The `v8.3.31` release of Ultralytics introduces enhancements to automatic batch size estimation during model training, which aims to optimize memory usage and manage CUDA memory issues more effectively.

📊 Key Changes
- **Batch Size Optimization**: Implemented `auto_batch` functionality to determine the best batch size by evaluating memory consumption.
- **Improved Profiling**: The profiling tools have been updated to include a `max_num_obj` parameter for better batch size accuracy.
- **Error Management**: Introduced logging for CUDA out-of-memory warnings and an automatic switch to CPU computation when necessary.
- **Documentation Updates**: Removed the `verbose` argument from training documentation as it was deemed ineffective.

🎯 Purpose & Impact
- **Efficient Memory Use**: Automatically adjusting batch sizes helps prevent overloading GPU memory, resulting in more efficient and stable training sessions. This is particularly beneficial for preventing abrupt interruptions due to memory errors.
- **Greater Reliability**: By switching to CPU processing when encountering memory errors, the system maintains training continuity, avoiding crashes and ensuring an uninterrupted user experience.
- **Simplified User Experience**: Streamlining training configuration by removing unnecessary options enhances usability, making the setup less complex for users.

What's Changed
* Remove `verbose` arg from train docs. by Y-T-G in https://github.com/ultralytics/ultralytics/pull/17257
* `ultralytics 8.3.31` add `max_num_obj` factor for `AutoBatch` by Laughing-q in https://github.com/ultralytics/ultralytics/pull/17514


**Full Changelog**: https://github.com/ultralytics/ultralytics/compare/v8.3.30...v8.3.31

📊 Key Changes
- **Memory Management**: Implemented a CPU fallback mechanism for task alignment calculations if a GPU `torch.OutOfMemoryError` occurs.
- **Method Refactoring**: Introduced a helper method `_forward` to elegantly manage memory overflow conditions.
- **Docker and Documentation Fixes**: Improved Docker image tagging and fixed a broken Jetson device documentation link.
- **Enhanced Features**: Simplified documentation examples and introduced a new `RegionCounter` module for easier region-based object counting.

🎯 Purpose & Impact
- **Stability and Reliability**: By ensuring task alignment processes can run on CPU under memory pressure, the update prevents application crashes and allows users with limited GPU resources to continue operations smoothly.
- **User Experience**: The changes make YOLO operations more flexible and robust, especially in environments with constrained computational resources, helping users to maintain performance without interruptions.
- **Documentation and Usability**: Improved documentation clarity makes it easier for both new and existing users to implement video analytics and other YOLO model features effectively. The `RegionCounter` addition simplifies integrating real-time object counting in specific video regions, broadening the tool's practical applications. 🔧

These updates and enhancements ensure that users have a smoother and more reliable experience with Ultralytics YOLO, particularly in resource-constrained settings.

What's Changed
* Fix Docker `jupyter` image naming by ambitious-octopus in https://github.com/ultralytics/ultralytics/pull/17479
* Include FPs for images with no labels in confusion matrix by Y-T-G in https://github.com/ultralytics/ultralytics/pull/17481
* Simplify Solutions Docs code examples by RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/17493
* Fix broken Jetson doc URL by lakshanthad in https://github.com/ultralytics/ultralytics/pull/17519
* Update models.md by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/17525
* Add return check for 'yolo solutions help' by RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/17518
* Update sony-imx500.md by ServiAmirPM in https://github.com/ultralytics/ultralytics/pull/17491
* Fix `file_name` in `save_crop` by M3nxudo in https://github.com/ultralytics/ultralytics/pull/17499
* Add region counter as ultralytics solution by RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/17439
* `ultralytics 8.3.30` run TAL on CPU if `torch.OutOfMemoryError` by Laughing-q in https://github.com/ultralytics/ultralytics/pull/17515

New Contributors
* ServiAmirPM made their first contribution in https://github.com/ultralytics/ultralytics/pull/17491
* M3nxudo made their first contribution in https://github.com/ultralytics/ultralytics/pull/17499

**Full Changelog**: https://github.com/ultralytics/ultralytics/compare/v8.3.29...v8.3.30