Ultralytics

🌟 Summary
The `v8.3.12` release of Ultralytics emphasizes the introduction of multi-point prompt support for the SAM and SAM2 models, allowing more flexible user interactions with segmentation tasks.

📊 Key Changes
- Introduced the ability to use multiple points and labels as prompts for a single object in SAM models. Points can now be structured in `[num_objects, num_points, 2]` for positions and `[num_objects, num_points]` for labels.
- Enhanced the analytics documentation and utilitarian functions for clearer data visualization and analysis.
- Deprecated the Ultralytics Explorer in favor of the Ultralytics HUB platform, consolidating tools for data exploration in a more unified and user-friendly environment.
- Optimized EdgeTPU exports to speed up compilation and improve configuration settings.

🎯 Purpose & Impact
- **Flexibility and Precision**: The support for multi-point prompts allows more detailed and complex object segmentation, providing better control and specificity for users dealing with intricate images.
- **Improved Usability**: The analytics update enhances user experience by simplifying the setup process, thereby benefiting both novices and experienced developers in understanding and utilizing data analytics functions.
- **Streamlined Experience**: Transition to Ultralytics HUB unifies resources, offering enhanced features and a simplified workflow for data exploration without the need for extensive coding, making it more accessible to a broader audience.
- **Efficiency Boost**: The EdgeTPU enhancements save time in model deployments, offering deeper debugging insights and faster operation, crucial for high-paced development environments.

What's Changed
* Update `analytics` solution by RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/16823
* Remove `explorer` Integration by RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/16842
* Reduce EdgeTPU to 3 compiler steps by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/16880
* `ultralytics 8.3.12` SAM and SAM2 multi-point prompts by Y-T-G in https://github.com/ultralytics/ultralytics/pull/16643


**Full Changelog**: https://github.com/ultralytics/ultralytics/compare/v8.3.11...v8.3.12

🌟 Summary
The release of Ultralytics YOLO `v8.3.11` focuses on a model naming overhaul and version synchronization for enhanced clarity.

📊 Key Changes
- 📦 Version bump from 8.3.10 to 8.3.11.
- 🔄 Model file names updated from `yolov8n` to `yolo11n` across varying export formats.
- ⚙️ Suppression of `onnx2tf` INT8-FP16 activation errors, ensuring smoother conversion processes.

🎯 Purpose & Impact
- **Naming Consistency:** By aligning model names across exports, users gain a more cohesive understanding of the model variation and version they are using.
- **User Convenience:** The streamlined naming aids in identifying and using model exports effectively, leading to a smoother workflow for developers.
- **Error Reduction:** Addressing specific activation errors enhances the reliability and usability of the exports, particularly when using different formats for deployment.

What's Changed
* Add HUB-SDK badges to CI Docs by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/16855
* Trigger GitHub Action on successful Docker Deploy by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/16858
* Only trigger additional Actions on new Ultralytics releases by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/16859
* Fix CI docker.yaml by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/16869
* Update YOLOv8n to YOLO11n in args by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/16873
* Add camera device to inference sources by Y-T-G in https://github.com/ultralytics/ultralytics/pull/16866
* `ultralytics 8.3.11` suppress `onnx2tf` INT8-FP16 act error by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/16874


**Full Changelog**: https://github.com/ultralytics/ultralytics/compare/v8.3.10...v8.3.11

🌟 Summary
The v8.3.10 release introduces support for HEIC images in predictions, which are commonly used by Apple devices. 📷

📊 Key Changes
- **HEIC Image Support:** Code enhancements allow for HEIC image loading and processing in predictions.
- **Documentation Update:** Added notes specifying that HEIC support is for inference only.
- **Version Update:** Bumped version from 8.3.9 to 8.3.10 to incorporate these modifications.

🎯 Purpose & Impact
- **Enhanced Compatibility:** Users can now use HEIC images for inference, expanding the range of compatible image formats.
- **Improved Clarity:** Documentation improvements help users understand the capabilities and limitations of HEIC support, ensuring a better user experience.

Additional updates in this release included documentation enhancements, improved image filename handling, and updates to ensure consistency in dataset names and integrative documentation. Each of these efforts contributes to smoother operation and better user clarity.

What's Changed
* Sort alphabetical integrations in docs by jk4e in https://github.com/ultralytics/ultralytics/pull/16819
* Fix empty paths while saving images with PIL bytes input by SheffeyG in https://github.com/ultralytics/ultralytics/pull/16809
* Fix typos in Tasks in docs so description matches example code by jk4e in https://github.com/ultralytics/ultralytics/pull/16825
* Fixed YOLO heads docstrings by ambitious-octopus in https://github.com/ultralytics/ultralytics/pull/16822
* Adding missing MNIST160 reference in Docs by jk4e in https://github.com/ultralytics/ultralytics/pull/16836
* `ultralytics 8.3.10` Apple iPhone HEIC support by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/16853


**Full Changelog**: https://github.com/ultralytics/ultralytics/compare/v8.3.9...v8.3.10

🌟 Summary
The latest release of Ultralytics, version 8.3.9, introduces detailed improvements to training log outputs and CSV handling, enhancing tracking and reporting capabilities for model training sessions.

📊 Key Changes
- **Version Update**: Incremented from 8.3.8 to 8.3.9.
- **Improved Training Logs**: Enhanced logging to record epochs and training durations more effectively.
- **CSV Enhancements**: Training time is now included in CSV metric outputs, enabling easier analysis.
- **batch Size Adjustment**: Dynamic adjustment of batch sizes for GPUs with over 16GB of memory.
- **Branding Updates**: Renamed default project and task identifiers from "YOLOv8" to "Ultralytics".
- **Editable Installations**: Removed editable (`-e`) pip install mode from continuous integration workflows.

🎯 Purpose & Impact
- 🕒 **Enhanced Time Tracking**: By logging training time in detail, users can better manage and evaluate the efficiency of their model training processes.
- 📈 **Better Metric Tracking**: Including training time in the CSV output simplifies performance tracking and comparative analysis across different runs.
- ⚡ **Optimized GPU Use**: Allowing larger batch sizes on high-memory GPUs enhances computational resource utilization, resulting in faster training and inference.
- 🎨 **Consistent Branding**: Unified project naming promotes a more cohesive brand identity across logging and monitoring platforms.
- 🧩 **Improved Code Maintenance**: Simplifying installations and configurations results in fewer potential errors during setup and better reliability of the development and deployment workflows.

What's Changed
* Rename loggers from YOLOv8 to Ultralytics by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/16784
* AutoBatch larger batches on larger GPUs by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/16785
* fix autobackend.py by lmk123568 in https://github.com/ultralytics/ultralytics/pull/16801
* Update `speed-estimation` solution by RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/16798
* Remove -e tag from CI by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/16797
* Fix wrong parameter count while profiling by Y-T-G in https://github.com/ultralytics/ultralytics/pull/16790
* `ultralytics 8.3.9` add training `time` to results CSV by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/16803

New Contributors
* lmk123568 made their first contribution in https://github.com/ultralytics/ultralytics/pull/16801

**Full Changelog**: https://github.com/ultralytics/ultralytics/compare/v8.3.8...v8.3.9

🌟 Summary
The v8.3.8 release of Ultralytics introduces performance optimizations by updating exception handling techniques throughout the codebase. 🚀

📊 Key Changes
- **Exception Handling Update**: Replaced `contextlib.suppress` with `try-except` blocks for faster exception handling.
- **Queue Management Code Enhancement**: Simplified and improved the queue management solution, making it more streamlined and user-friendly.
- **Environment Compatibility**: Improved image display logic for Colab and Kaggle environments.
- **Documentation Improvements**: Enhanced clarity and correctness of integration docs with Weights & Biases.

🎯 Purpose & Impact
- **Performance Improvement**: The use of `try-except` over `contextlib` is significantly faster (up to 4x), enhancing the application's performance.
- **Code Readability and Maintenance**: Clear exception handling increases code transparency and eases debugging and maintenance for developers.
- **User Experience**: Improved handling of queue management logic ensures a smoother workflow for users, particularly in object tracking and counting scenarios.
- **Cross-Platform Compatibility**: Fixes related to image display boost the usability for Jupyter-based notebooks, including popular cloud platforms like Colab and Kaggle.
- **Up-to-date Version**: Version bump to 8.3.8 ensures users benefit from all latest improvements and bug fixes.

What's Changed
* Fix synthetic COCO for missing labels by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/16748
* Add package name to YOLO checks errors by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/16770
* Prevent final val if val=False by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/16776
* Fix `results.show()` for Jupyter notebooks by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/16783
* Fix display of admonition in W&B docs by jk4e in https://github.com/ultralytics/ultralytics/pull/16777
* Update `queue-management` solution by RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/16772
* Update usage for custom Predictor by Laughing-q in https://github.com/ultralytics/ultralytics/pull/16771
* Respect ENV value for `OMP_NUM_THREADS` if set by Burhan-Q in https://github.com/ultralytics/ultralytics/pull/16751
* `ultralytics 8.3.8` replace `contextlib` with `try` for speed by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/16782


**Full Changelog**: https://github.com/ultralytics/ultralytics/compare/v8.3.7...v8.3.8

🌟 Summary
The v8.3.7 release introduces several enhancements, with a focus on fixing argument handling for custom predictors in YOLO, along with updates for performance and stability improvements.

📊 Key Changes
- **Custom Predictor Argument Fix**: Corrected the usage of arguments in `model.predict()` to support custom predictors.
- **Docker Image Update**: The base Docker image is now upgraded to utilize PyTorch 2.4.1, with CUDA 12.1 and cuDNN 9.
- **New Script for Synthetic Datasets**: Added a function to create synthetic COCO datasets, supporting data testing and augmentation.
- **Enhanced AutoBatch Memory Management**: Improved GPU memory handling during autobatching to optimize resource usage.
- **Added `OMP_NUM_THREADS=1`**: Adjusted Docker configurations for improved CPU management and performance.

🎯 Purpose & Impact
- 🛠 **Improved Custom Predictor Functionality**: Ensures smoother operation for users implementing custom predictors, providing accurate argument handling on initialization.
- 🚀 **Performance Boost**: The Docker update incorporates the latest improvements in PyTorch, enhancing model training and inference speed.
- 🌟 **Simplified Data Handling**: The synthetic COCO dataset script facilitates the creation of testing datasets, helping developers easily prepare data for model validation.
- 🧹 **Efficient Resource Management**: Changes in autobatch memory handling and Docker configuration aim to reduce memory overhead and CPU saturation, ensuring better performance on a variety of systems.
- 📈 **Streamlined Development**: These updates collectively enhance the developer experience by optimizing workflows, reducing potential bugs, and improving code clarity.

What's Changed
* FROM pytorch/pytorch:2.4.1-cuda12.1-cudnn9-runtime by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/16725
* Fix absolute stylesheet CSS path by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/16726
* New `create_synthetic_coco_dataset` function by UltralyticsAssistant in https://github.com/ultralytics/ultralytics/pull/16742
* Added YOLO11 to Explorer models by ambitious-octopus in https://github.com/ultralytics/ultralytics/pull/16739
* Fix NVIDIA Typo by lakshanthad in https://github.com/ultralytics/ultralytics/pull/16728
* setuptools>=70.0.0 fix CVE-2024-6345⁠ by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/16727
* AutoBatch improve cache clearing by UltralyticsAssistant in https://github.com/ultralytics/ultralytics/pull/16744
* Adding `OMP_NUM_THREADS=1` by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/16747
* `ultralytics 8.3.7` custom Predictor args fix by Rick-v-E in https://github.com/ultralytics/ultralytics/pull/16734


**Full Changelog**: https://github.com/ultralytics/ultralytics/compare/v8.3.6...v8.3.7