Ultralytics

🌟 Summary
The `v8.3.63` release focuses on robustness improvements, developer convenience, and enhanced edge-case handling through better `sudo` command detection, optimized imports, and model training consistency in distributed environments.

📊 Key Changes
- **Sudo Detection Utility**: Added `is_sudo_available()` function to streamline installation in export processes (e.g., Edge TPU, IMX500).
- **Optimized Imports**: Simplified `thop` and other imports to improve loading efficiency for developers.
- **Distributed Training Fix**: Corrected learning rate inconsistencies between single-GPU and DDP training by reapplying model attributes.
- **Documentation Enhancements**: Improved link and file consistency (e.g., underscores to hyphens in filenames), and clarified version references for testing.
- **Dataloader Cleanup**: Prevented errors during worker shutdown for cases without initialized workers.

🎯 Purpose & Impact

- **Improved Stability**:
- 🛠️ Systems without `sudo` now handle export dependencies more smoothly, reducing user setup issues.
- 🚀 DDP training now avoids unintended fallback values, ensuring consistent performance across all setups.

- **Enhanced Developer Experience**:
- ⚡ Faster module loading due to scoped imports and redundant checks removal.
- 📚 Clearer and more accessible documentation aids both developers and end-users in understanding workflows.

- **Error Prevention**:
- ✅ Edge-case safeguards for dataloaders avoid crashes during cleanup tasks.

This update reflects thoughtful optimizations for stability, usability, and performance while addressing minor bugs and maintaining interoperability. 🚀

What's Changed
* Update `8.3.82` with `8.2.82` in `sam-2.md` by RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/18712
* Simplify `thop` imports by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/18717
* Fix automatic optimizer LR with DDP training by Y-T-G in https://github.com/ultralytics/ultralytics/pull/18641
* Update HUB alt text by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/18719
* Fix dataloader cleanup error with no workers by Y-T-G in https://github.com/ultralytics/ultralytics/pull/18732
* `ultralytics 8.3.63` IMX500 sudo install fix by ambitious-octopus in https://github.com/ultralytics/ultralytics/pull/18714


**Full Changelog**: https://github.com/ultralytics/ultralytics/compare/v8.3.62...v8.3.63

---

📊 Key Changes
- **Fix for Non-Deterministic Transforms**: Resolved randomness issues in data augmentation with `albumentations>=1.4.21` by adding support for setting a random seed. 🔧
- **Workflow and Documentation Enhancements**:
- Renamed GitHub workflow files for consistency (`.yaml` → `.yml`). 📂
- Updated project licensing headers for clarity across all source files and configurations. 📝
- Refreshed metadata to display the current year (2025) in documentation. 📅
- **Bug Fixes**: Addressed sporadic dataloader freezes during consecutive training sessions. 🛠️
- **Code Clean-Up**: Streamlined hyperparameter mutation logic by reducing unnecessary data access. ✨

---

🎯 Purpose & Impact
- **Consistent Training Results**: Deterministic transformations allow reproducible results in model training, improving debugging and performance evaluation. 📈
- **Improved Usability**: Updated workflows and file organization ensure better developer experience and easier maintenance for contributors. 🧑‍💻
- **Enhanced Stability**: Fixes to the dataloader and optimization pipeline enhance reliability for users running training sessions repeatedly. 🚦
- **Professional Branding**: Revised licensing headers and metadata maintain a polished and up-to-date project representation. 🌐

⚙️ Whether you're a developer improving AI systems or a researcher fine-tuning models, this release ensures smoother, more predictable processes while adhering to modern software conventions. 🎉

What's Changed
* Consistent workflow file suffix *.yml by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/18668
* `ci.yml` and `docker.yml` rename by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/18671
* Fix MNN example missing BGR to RGB conversion by jules-ai in https://github.com/ultralytics/ultralytics/pull/18689
* Remove unused dict values in `items()` with `values()` by Kayzwer in https://github.com/ultralytics/ultralytics/pull/18651
* Update 2024 to 2025 by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/18695
* Standardize license headers in Python files by pderrenger in https://github.com/ultralytics/ultralytics/pull/18696
* Fix random dataloader freezes by Y-T-G in https://github.com/ultralytics/ultralytics/pull/18697
* Improve headers and comments in TOML/YAML files by pderrenger in https://github.com/ultralytics/ultralytics/pull/18698
* `ultralytics 8.3.62` Fix non-deterministic transforms with `albumentations>=1.4.21` by Y-T-G in https://github.com/ultralytics/ultralytics/pull/18701

New Contributors
* jules-ai made their first contribution in https://github.com/ultralytics/ultralytics/pull/18689

**Full Changelog**: https://github.com/ultralytics/ultralytics/compare/v8.3.61...v8.3.62

🌟 Summary
The `v8.3.61` release restores compatibility with Python 3.8, refines several utility components, and makes user-facing improvements for better prediction workflows. It smooths the usability, performance, and flexibility of the Ultralytics ecosystem. 🚀💡

---

📊 Key Changes
- **Python 3.8 Compatibility Restored**: Ensures older Python versions, including 3.8, are now supported by replacing operations incompatible with earlier versions. 🐍✅
- Replaced `|` operator for dictionary merging with the Python 3.8-compatible `**` method.
- **Prediction API Simplification**: The `Predictor` and `SAM2Predictor` classes now output results in a consolidated object (`result`), replacing the previous multi-output format (`masks, scores, boxes`).
- Example updates reflect this newer, simpler API. 🧰
- **CI Workflow Improvements**: Adjusted triggers and settings in GitHub Actions workflows for smoother testing and CI processes.
- **Minor Bug Fixes**: Addressed issues in utility functions, prediction methods, and loss calculations to improve overall reliability and prevent unexpected errors. 🛠️
- **Version Bumped to 8.3.61**: Reflects these refinements in the package version.

---

🎯 Purpose & Impact
- **Broader Python Compatibility**: Expands usability for projects running Python 3.8, accommodating users on older infrastructure without losing core functionality. 🌎
- **Simplified Model Predictions**: A single-output format for `Predictor` and `SAM2Predictor` reduces user friction, improves code readability, and simplifies integration into pipelines. Especially helpful for new users! 🧩🚀
- **Improved Reliability**: Minor bug fixes enhance stability, ensuring smoother performance across diverse use cases.
- **CI Stability & Testing Precision**: Improvements to workflow configurations ensure better support for continuous integration and testing scenarios.

---

Note for Users 🤓:
If you’ve been using the `Predictor` or `SAM2Predictor` classes, make sure to update your scripts with the new `result` structure (e.g., use `result.masks`, `result.boxes`, etc.) instead of relying on separate outputs. This alignment will make your workflows cleaner and future-proof! 😊

What's Changed
* Fix broken examples in SAM Predictor docstrings by Y-T-G in https://github.com/ultralytics/ultralytics/pull/18665
* `ultralytics 8.3.61` Restore Python 3.8 compatibility by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/18666


**Full Changelog**: https://github.com/ultralytics/ultralytics/compare/v8.3.60...v8.3.61

🌟 Summary
This update primarily fixes CoreML segmentation output handling, improves documentation, and enhances the usability of model features for developers and end users. 🔄✨

---

📊 Key Changes
- **CoreML Segmentation Fix**: Improved logic for processing segmentation model outputs in `autobackend.py` (fixed reverse order issue for specific use cases).
- **Docker Update**: Dockerfile upgraded to PyTorch `2.5.1` (with CUDA 12.4 and cuDNN 9), enabling improved compatibility and performance for Docker-based workflows. 🐳⚡
- **Colab Integrations**: Added Colab badges to various documentation pages for easy, hands-on experimentation with datasets and tutorials. 📚🔗
- **Enhanced Auto-Annotation Documentation**: Updated guides for segmentation auto-annotation, adding clarity around supported models and parameter configurations. 🖼️✅
- **Bug Reporting Improvements**: Adjusted GitHub issue templates to request full traceback info for better debugging efficiency. 🛠️🔍
- **Standardized String Formatting**: Converted strings to consistently use double-quoted f-strings for better code clarity and maintainability. 🖊️

---

🎯 Purpose & Impact

- **CoreML Update**:
- 🛠 **Purpose**: Fix and streamline CoreML model support, specifically for segmentation outputs.
- 🌟 **Impact**: Smoother deployment for Apple-device-specific workflows with reduced risk of errors in segmentation processing.

- **Docker Upgrade**:
- 🚀 **Purpose**: Ensure containerized environments stay up-to-date and performant with compatibility fixes.
- 🌟 **Impact**: Faster inference and training workflows with enhanced reliability.

- **Colab Additions**:
- 🛠 **Purpose**: Enable effortless model experimentation with interactive tools directly from the documentation.
- 🌟 **Impact**: Lowers the entry barrier for new users while improving developer productivity.

- **Auto-Annotation Improvements**:
- 🎯 **Purpose**: Clarify how to use segmentation models like SAM and MobileSAM for large datasets.
- 🌟 **Impact**: Saves time in dataset labeling by simplifying setup and enabling quick-start options.

- **Standardized String Formatting**:
- 🖊️ **Purpose**: Improve code readability and ease of maintenance for developers.
- 🛡 **Impact**: Cleaner, more professional code with improved developer experience.

- **Bug Reporting Guidelines**:
- 🚨 **Purpose**: Collect more detailed user environment data to speed up issue resolution.
- 🌍 **Impact**: Quicker turnaround in fixing bugs due to detailed diagnostic info.

No breaking changes in this release, ensuring smooth upgrades across workflows! 🛡💡

What's Changed
* Apply Ruff 0.9.0 by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/18622
* Add new Colab Notebooks badges to Docs by RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/18575
* Apply `ruff==0.9.0` formatting by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/18624
* Update `val.md` by RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/18645
* Update issue templates with better instructions by Y-T-G in https://github.com/ultralytics/ultralytics/pull/18346
* Dockerfile FROM pytorch/pytorch:2.5.1-cuda12.4-cudnn9-runtime by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/18650
* Add warning for `task=classify` with `mode=track` by RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/18621
* Warn and set `task=detect` and `mode=track` for `task=track` by RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/18620
* Add `mobile-sam` auto-annotation to segmentation datasets docs by RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/18654
* `ultralytics 8.3.60` Fix CoreML Segment inference by Y-T-G in https://github.com/ultralytics/ultralytics/pull/18649


**Full Changelog**: https://github.com/ultralytics/ultralytics/compare/v8.3.59...v8.3.60

🌟 Summary
The latest release, `v8.3.59`, introduces the ability to load any `torchvision` model as a backbone, along with several quality-of-life updates, including enhanced Docker support, dataset path refinements, and usability improvements in documentation and tools. 🚀

---

📊 Key Changes
- 🔥 **Custom TorchVision Backbone Support**: Allows integration of any `torchvision` model (e.g., EfficientNet, MobileNet, ResNet) as YOLO backbones! Includes options for pretrained weights and layer customization.
- 🖼️ **Expanded Segmentation Mask Support**: Added compatibility for `.jpg` masks alongside existing `.png` support.
- 🐛 **Validation Enhancements for INT8 Calibration**: New checks ensure calibration datasets meet batch size requirements, providing more robust error handling.
- 🛠️ **Improved Docker Environment**: Simplified JupyterLab installations and introduced retry mechanisms for Docker image pushes.
- 🔧 **Updated Dataset Paths**: Refined YAML dataset path structures for better organization and reduced misconfigurations.
- ⚙️ **Enhanced Multi-Processing Documentation**: Help added for common Windows-related training errors (e.g., `RuntimeError`) with clear solutions.
- 📊 **New Benchmarks**: Extended NVIDIA DeepStream and Coral TPU performance benchmarks for development on Jetson devices and Raspberry Pi (including Pi 5).

---

🎯 Purpose & Impact
- **Flexibility & Power with TorchVision Backbones**:
- Users can now integrate models like ConvNext and MobileNet directly into YOLO pipelines.
- Pretrained weights accelerate training for both object detection and classification tasks. 🔄
- **Streamlined Segmentation Workflows**:
- Compatibility with `.jpg` masks eliminates a frequent need for manual file conversions, saving time. 🕒
- **INT8 Improvements**:
- The validation on calibration size prevents breakdowns in compression workflows, ensuring higher-quality deployment setups.
- **Smoother Docker & DevOps**:
- Better Docker resilience and JupyterLab setup reduce installation friction for developers. 🐳
- **Training Guidance on Windows**:
- Clear troubleshooting advice mitigates pitfalls for users launching scripts in Windows environments for seamless training experiences.
- **Enhanced Benchmark Documentation**:
- Developers can now better choose the hardware and YOLO model precision (e.g., FP32, FP16, or INT8) for NVIDIA Jetson or Edge TPU use cases. 📈

---

This release offers powerful new capabilities for model customization and smoother workflows, making it a significant upgrade for developers working with YOLO and associated tools. 🎉

What's Changed
* Add instructions to enable W&B logging by Y-T-G in https://github.com/ultralytics/ultralytics/pull/18546
* Add warning about Windows multi-processing error when launching training by Y-T-G in https://github.com/ultralytics/ultralytics/pull/18547
* Ultralytics Refactor https://ultralytics.com/actions by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/18555
* Use uv for Dockerfile-jupyter by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/18567
* Add retries to Docker pushes by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/18565
* Update Benchmarks for NVIDIA DeepStream running on NVIDIA Jetson by lakshanthad in https://github.com/ultralytics/ultralytics/pull/18603
* Add benchmarks for Pi 4B/5 by Skillnoob in https://github.com/ultralytics/ultralytics/pull/18580
* Update `package-seg.yaml` by RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/18594
* Verify dataset >= batch size on INT8 export calibration by Y-T-G in https://github.com/ultralytics/ultralytics/pull/18611
* Fix incorrect docstring for bbox_iou function by visionNoob in https://github.com/ultralytics/ultralytics/pull/18579
* Include .jpg in mask converter by Y-T-G in https://github.com/ultralytics/ultralytics/pull/18576
* `ultralytics 8.3.59` Add ability to load any `torchvision` model as module by Y-T-G in https://github.com/ultralytics/ultralytics/pull/18564

New Contributors
* visionNoob made their first contribution in https://github.com/ultralytics/ultralytics/pull/18579

**Full Changelog**: https://github.com/ultralytics/ultralytics/compare/v8.3.58...v8.3.59

🌟 Summary
The `v8.3.58` release introduces an update to the YOLO model benchmarking utility for TensorRT, documentation enhancements, and new features to improve usability and performance for developers and users. 🚀🛠️

---

📊 Key Changes
- **TensorRT Model Benchmarking Improvement**:
- Updated benchmarking to use `uint8` (integer) input data instead of `float32` (decimals) for classification tasks, reflecting real-world input formats.
- **Documentation Enhancements**:
- Embedded instructional videos in object counting and model exporting guides for clarity. 🎥
- Updated integration documentation for YOLO11, replacing mentions of YOLOv8.
- **New Training Argument**:
- Added `multi_scale` training option in documentation for dynamic image sizes during training. 🌈
- **Docker Optimization**:
- Added a `.dockerignore` file to exclude unnecessary files, improving build efficiency and security.

---

🎯 Purpose & Impact
- **Purpose**:
- Optimize benchmarking processes and align input data with typical formats for more accurate performance evaluations during TensorRT model testing.
- Enhance usability through better instructional resources and accurate documentation for new model versions.
- Introduce dynamic training options to improve model adaptability for various image sizes.
- Improve Docker image builds by reducing clutter and improving security. 🔒

- **Impact**:
- TensorRT users will benefit from faster and more realistic benchmarks for classification tasks. 🏎️
- Documentation updates simplify learning and onboarding for both new and advanced users. 📘
- Developers now have the option to train models across multiple resolutions, potentially enhancing inference accuracy.
- Docker environments become leaner and more secure, supporting cleaner deployments. 🐋

This release is an essential step forward for developers seeking both practical performance boosts and improved clarity in documentation! 💡

What's Changed
* Add https://youtu.be/K69DUpSBNdA to docs by RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/18507
* Update `yolov8` to `yolo11` in `tensorrt.md` by RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/18513
* Add `multi_scale` training argument to docs by Y-T-G in https://github.com/ultralytics/ultralytics/pull/18531
* Create .dockerignore by glenn-jocher in https://github.com/ultralytics/ultralytics/pull/18534
* `ultralytics 8.3.58` Use `uint8` type for TensorRT Profile by Laughing-q in https://github.com/ultralytics/ultralytics/pull/18327


**Full Changelog**: https://github.com/ultralytics/ultralytics/compare/v8.3.57...v8.3.58