Ultralytics

🌟 Summary
This release standardizes the folder structure for datasets in Ultralytics, making dataset paths more consistent and user-friendly across multiple YAML configuration files. 📁✨

📊 Key Changes
- Updated the version to 8.3.165.
- Unified dataset image paths in all relevant YAML files to follow a consistent structure: `images/train`, `images/val`, and `images/test`.
- Updated comments to reflect minor changes in dataset download sizes.

🎯 Purpose & Impact
- 🧩 **Consistency:** All datasets now use the same folder structure, reducing confusion and minimizing setup errors for users.
- 🚀 **Ease of Use:** Switching between datasets or automating workflows is now simpler, as you no longer need to adjust image paths for different datasets.
- 🛠️ **Better Maintenance:** This change streamlines future updates and documentation, making Ultralytics more approachable for both new and experienced users.

Overall, this update improves the reliability and usability of dataset management, helping users work more efficiently with Ultralytics models and tools.

What's Changed
* `ultralytics 8.3.165` Update datasets YAML's by RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/21353


**Full Changelog**: https://github.com/ultralytics/ultralytics/compare/v8.3.164...v8.3.165

🌟 Summary
This release delivers a critical fix to YOLO detection validation metrics, improves dataset flexibility, enhances export reliability, and polishes documentation and developer experience. 🚀

📊 Key Changes

- **Metrics Fix in YOLO Detection Validation**: Corrected the assignment of mAP50 and mAP50-95 metrics, ensuring accurate reporting in validation results.
- **Flexible Text Sample Handling in Datasets**: Added a `max_samples` parameter for controlling the number of text samples in `GroundingDataset` and improved logic for negative text selection.
- **Export Improvements & Warnings**: Enhanced TensorRT export logic with better device handling and warnings for dynamic batch sizes, helping users avoid common pitfalls.
- **Classification Dataset Compatibility**: Now supports `valid/` as a fallback validation folder, improving compatibility with Roboflow and similar dataset exports.
- **Type Annotation Consistency**: Standardized all references from `numpy.ndarray` to `np.ndarray` across code and documentation for clarity.
- **HEIC Image Support Update**: Switched from `pillow_heif` to `pi-heif` for HEIC image decoding, simplifying license compliance.
- **Documentation & UI Enhancements**:
- Embedded a new YouTube video in the data annotation guide for easier learning.
- Improved object counting example and clarified usage of the `sweep_annotator` method.
- Removed unnecessary CSS animations from the Similarity Search web page for a faster, cleaner interface.
- **Minor Fixes**:
- Corrected an error message in classification augmentations for better developer clarity.

🎯 Purpose & Impact

- **Accurate Model Evaluation**: Ensures that users and researchers see the correct mAP metrics, preventing confusion and supporting trustworthy model comparisons.
- **Greater Dataset Flexibility**: Developers can fine-tune text augmentation, leading to more efficient training and better use of resources.
- **Smoother Model Export**: Clearer export warnings and device handling reduce errors and improve deployment reliability, especially for advanced users exporting to TensorRT.
- **Broader Dataset Compatibility**: Makes it easier to use datasets from popular tools like Roboflow without manual folder renaming.
- **Improved Developer Experience**: Consistent type hints and clearer error messages make the codebase more approachable and maintainable.
- **Better Documentation & Usability**: New video resources and UI tweaks help both new and experienced users get started and work more efficiently.
- **License Compliance**: The change in HEIC image support ensures continued open-source compliance and easier integration in commercial projects.

---

This update is recommended for all users, especially those validating detection models, working with custom datasets, or exporting models for deployment. 🚀🛠️

What's Changed
* Fix inconsistent `max_samples` value between `GroundingDataset` and `YOLOMultiModalDataset` by Y-T-G in https://github.com/ultralytics/ultralytics/pull/21286
* Add max batch warning for dynamic model with NMS by Y-T-G in https://github.com/ultralytics/ultralytics/pull/21320
* Support `valid/` as fallback validation folder for classification datasets by JamesBond6873 in https://github.com/ultralytics/ultralytics/pull/21321
* Fix wrong error message in `classify_augmentations` by Toprak2 in https://github.com/ultralytics/ultralytics/pull/21322
* Update Python hints for augmentation functions by Laughing-q in https://github.com/ultralytics/ultralytics/pull/21329
* Eliminate extra `CSS` from `similarity-search.html` by RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/21335
* Update `sweep_annotator` example by RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/21331
* Switch `pillow_heif` to `pi_heif` by picsalex in https://github.com/ultralytics/ultralytics/pull/21339
* Add https://youtu.be/iBk6S-PHwS0 to docs by RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/21346
* `ultralytics 8.3.164` Fix swapped mAP50 and mAP50-95 in COCOEval stats by Y-T-G in https://github.com/ultralytics/ultralytics/pull/21350

New Contributors
* JamesBond6873 made their first contribution in https://github.com/ultralytics/ultralytics/pull/21321
* Toprak2 made their first contribution in https://github.com/ultralytics/ultralytics/pull/21322

**Full Changelog**: https://github.com/ultralytics/ultralytics/compare/v8.3.163...v8.3.164

🌟 Summary
This release brings smarter label validation, improved region counting, easier video frame management, and enhanced export and CI workflows for a smoother and more reliable Ultralytics experience. 🚀✨

📊 Key Changes

- **Label Verification Tolerance** 🏷️
Added a 1% tolerance to label normalization checks, reducing false errors from tiny rounding issues in datasets.
- **Region Counter Overhaul** 🔢
Improved the RegionCounter solution for easier setup and management of multiple counting zones, with clearer region drawing and more robust counting logic.
- **Video Frame Naming Fixes** 🎬
Optimized how video frames and prediction images are saved, ensuring consistent and intuitive file naming—especially for videos with complex filenames.
- **YOLOE Model Fusion Enhancement** 🤖
Refined model fusion to better support YOLOE models with positional encoding, improving deployment reliability.
- **ONNX Export Dependency Updates** 📦
Updated ONNXslim to version 0.1.59 for better model export optimization, then streamlined export requirements by removing unnecessary dependencies.
- **Documentation Upgrade** 🎥
Embedded a YouTube tutorial in the validation docs, making it easier to learn how to export validation results in various formats.
- **Continuous Integration (CI) Improvements** 🛠️
- JetPack tests now run all jobs regardless of failures, providing more complete test results.
- Added NVIDIA Jetson hardware checks and improved Slack notifications for CI failures, ensuring faster feedback and better hardware compatibility.

🎯 Purpose & Impact

- **For All Users:**
- Fewer frustrating false errors during data validation, especially when using datasets from different sources.
- Easier and more reliable management of video predictions and region-based object counting.
- Smoother experience exporting and optimizing models for deployment, with up-to-date dependencies and fewer installation issues.
- More accessible documentation and learning resources, thanks to embedded video guides.

- **For Developers & Advanced Users:**
- Improved CI workflows mean faster detection and resolution of issues, especially for Jetson and edge device deployments.
- Enhanced model fusion and export logic supports a broader range of architectures and deployment scenarios.

Overall, this update makes Ultralytics tools more robust, user-friendly, and ready for diverse real-world applications! 🌍💡

What's Changed
* Add https://youtu.be/zHxwDkYShNc to docs by RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/21287
* Skip `region` initialization every frame `~2x` faster by RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/21288
* YOLOE: Fix model text embedding fusing by xifeng0126 in https://github.com/ultralytics/ultralytics/pull/21277
* Update JetPack test matrix to run all jobs regardless of failure by lakshanthad in https://github.com/ultralytics/ultralytics/pull/21290
* Add Jetson CI for Slack notifications on failure by lakshanthad in https://github.com/ultralytics/ultralytics/pull/21291
* Refactor Summary job in CI for notifications by lakshanthad in https://github.com/ultralytics/ultralytics/pull/21293
* Optimize the name of video file and frames when using `save_txt` and `save_frames` by jugal-sheth in https://github.com/ultralytics/ultralytics/pull/21276
* Add `onnxslim>=0.1.59` to TOML export dependencies by inisis in https://github.com/ultralytics/ultralytics/pull/21302
* Remove `onnxslim` from `pyproject.toml` due to Jetson6 docker tests failing by lakshanthad in https://github.com/ultralytics/ultralytics/pull/21304
* `ultralytics 8.3.163` Add `1%` tolerance for labels normalization check by Laughing-q in https://github.com/ultralytics/ultralytics/pull/21310

New Contributors
* jugal-sheth made their first contribution in https://github.com/ultralytics/ultralytics/pull/21276
* xifeng0126 made their first contribution in https://github.com/ultralytics/ultralytics/pull/21277

**Full Changelog**: https://github.com/ultralytics/ultralytics/compare/v8.3.162...v8.3.163

🌟 Summary
This release (v8.3.162) brings improved reliability and consistency to model loading, enhanced hardware compatibility, and several quality-of-life updates for both developers and users. 🚀🛠️

📊 Key Changes

- **Standardized Model Loading:**
All direct uses of `torch.load` are replaced with Ultralytics' `torch_load` utility, ensuring consistent and robust model file handling throughout the codebase.
- **Improved Device Compatibility:**
Loading of cached text embeddings in YOLOE and YOLO-World now explicitly supports both CPU and GPU, preventing device mismatch errors.
- **Intel Hardware Detection:**
Added a new utility to detect Intel CPUs and GPUs, allowing Ultralytics tools to recommend OpenVINO exports for optimal performance on Intel hardware.
- **Enhanced Metrics Plotting:**
Metric plots for detection, segmentation, and pose tasks now include clearer labeling and improved processing, making evaluation results easier to interpret.
- **Relative Path Support for Grounding Datasets:**
Open-vocabulary model training now supports relative dataset paths, simplifying custom dataset management.
- **CopyPaste Augmentation Fix:**
The CopyPaste augmentation no longer modifies original images in-place, preserving data integrity during training.
- **Dependency Version Pinning:**
The `ai-edge-litert` package is now pinned to versions `>=1.2.0,<1.4.0` to ensure stable TensorFlow SavedModel exports.
- **Optional Typing Stubs:**
Introduced an optional dependency group for typing stubs, improving code completion and static analysis for developers.
- **Assorted Bug Fixes:**
Includes fixes for open-vocabulary evaluation, dataset handling, and minor code improvements.

🎯 Purpose & Impact

- **Reliability & Consistency:**
By standardizing model loading with `torch_load`, users and developers benefit from fewer bugs and more predictable behavior when working with PyTorch models.
- **Better Hardware Support:**
Users with Intel hardware now receive tailored export recommendations, and device-aware embedding loading prevents frustrating training errors.
- **Improved Usability:**
Clearer metric plots, support for relative dataset paths, and safer augmentations make the platform easier and safer to use for both new and advanced users.
- **Developer Experience:**
Optional typing stubs and codebase improvements enhance code quality, autocompletion, and maintainability.
- **Export Stability:**
Pinning dependencies reduces the risk of export failures, ensuring smoother model conversion workflows.

Overall, v8.3.162 delivers a more robust, user-friendly, and developer-friendly experience across the Ultralytics ecosystem! 🎉

What's Changed
* Optional dependency group for typing stubs by jorenham in https://github.com/ultralytics/ultralytics/pull/21137
* YOLOE: specify device for text embedding cache loading by Laughing-q in https://github.com/ultralytics/ultralytics/pull/21237
* Fix `is_lvis` check for open-vocabulary models evaluation by ImJaewooChoi in https://github.com/ultralytics/ultralytics/pull/21245
* Fix metrics plotting for `Pose` and `Segment` tasks by Laughing-q in https://github.com/ultralytics/ultralytics/pull/21259
* Support intel XPU discovery by ambitious-octopus in https://github.com/ultralytics/ultralytics/pull/21242
* Add `DATASET_DIR` relative path compatibility for grounding datasets by Laughing-q in https://github.com/ultralytics/ultralytics/pull/21256
* Avoid modifying original image during `CopyPaste` augmentation by Y-T-G in https://github.com/ultralytics/ultralytics/pull/21262
* Pin `ai-edge-litert>=1.2.0,<1.4.0` by Laughing-q in https://github.com/ultralytics/ultralytics/pull/21265
* `ultralytics 8.3.162` Replace `torch.load` calls with patched `torch_load` method that defaults to `weights_only=False` by Y-T-G in https://github.com/ultralytics/ultralytics/pull/21260

New Contributors
* jorenham made their first contribution in https://github.com/ultralytics/ultralytics/pull/21137

**Full Changelog**: https://github.com/ultralytics/ultralytics/compare/v8.3.161...v8.3.162

🌟 Summary
This release focuses on making dataset paths simpler and more consistent across Ultralytics, improving documentation, and enhancing compatibility for various platforms and users. 🗂️✨

📊 Key Changes

- **Dataset Path Simplification:** All dataset YAML files, documentation, and code examples now use direct folder names (like `coco`) instead of legacy relative paths (like `../datasets/coco`).
- **Documentation & Example Updates:** All guides, tutorials, and code snippets have been updated to reflect the new, cleaner dataset path format for detection, segmentation, pose, and OBB tasks.
- **Improved PaddlePaddle Support:** PaddlePaddle dependencies are now pinned for ARM64 devices, ensuring smoother installation and compatibility, especially on platforms like Raspberry Pi.
- **Deprecation Warning for Examples:** Added a clear notice that community-contributed examples will be retired in Ultralytics v8.4.0, encouraging users to rely on official documentation.
- **Sony IMX500 Benchmark Corrections:** Updated and clarified benchmark results and dataset naming for YOLO11n and YOLOv8n on Sony IMX500 devices.
- **Performance & Reliability Fixes:**
- Improved label file writing for the VisDrone dataset, making annotation conversion faster and more reliable.
- Corrected expected label counts for specific datasets to improve verification accuracy.
- **API & Documentation Enhancements:**
- Improved type annotations and documentation for result export and summary methods, making the API clearer for developers.
- Refactored VisualAISearch for simpler structure and more consistent logging.

🎯 Purpose & Impact

- **Easier Dataset Management:** Users can now specify datasets more intuitively, reducing confusion and setup errors, especially when using custom `datasets_dir` locations.
- **Better Cross-Platform Support:** ARM64 users (e.g., Raspberry Pi) will experience fewer installation issues with PaddlePaddle, broadening hardware compatibility.
- **Clearer Guidance:** Updated documentation and deprecation warnings help users find the most reliable, up-to-date resources, improving onboarding and reducing maintenance overhead.
- **Improved Reliability:** Fixes to dataset verification and annotation conversion mean fewer errors and smoother training experiences.
- **Developer Friendliness:** Enhanced type hints and API docs make it easier for developers to build on and integrate with Ultralytics tools.

---

This update is all about making Ultralytics easier to use, more robust, and ready for a wider range of users and devices! 🚀

What's Changed
* Fix the expected count for `final_mixed_train_no_coco_segm` in `GroundingDataset` by ImJaewooChoi in https://github.com/ultralytics/ultralytics/pull/21215
* Add deprecation warning to `ultralytics/examples` by Y-T-G in https://github.com/ultralytics/ultralytics/pull/21208
* Fix: Move YOLO label file writing outside annotation loop in `visdrone2yolo` function by banu4prasad in https://github.com/ultralytics/ultralytics/pull/21226
* Update Sony IMX Doc with COCO128 Benchmarks by lakshanthad in https://github.com/ultralytics/ultralytics/pull/21227
* Fix COCO128 capitalization in Sony IMX Doc by lakshanthad in https://github.com/ultralytics/ultralytics/pull/21229
* Pin `paddlepaddle==3.0.0` for ARM64 by Laughing-q in https://github.com/ultralytics/ultralytics/pull/21238
* Add `summary` method in `val.md` by RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/21236
* Remove `BaseSolution` inheritance in `VisualAISearch` by RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/21240
* `ultralytics 8.3.161` Eliminate `../datasets/` in data yaml files for better `datasets_dir` compatibility by Laughing-q in https://github.com/ultralytics/ultralytics/pull/21091

New Contributors
* ImJaewooChoi made their first contribution in https://github.com/ultralytics/ultralytics/pull/21215

**Full Changelog**: https://github.com/ultralytics/ultralytics/compare/v8.3.160...v8.3.161

🌟 Summary
This release brings improved keypoint handling, smarter data augmentation, and enhanced usability for training and exporting models, making YOLO workflows more robust and user-friendly. 🚀

📊 Key Changes
- **Keypoint Clipping for Visualization:** Keypoints are now clipped to stay within image boundaries, ensuring better training data quality and more accurate visualizations.
- **Simplified Data Augmentation Access:** Data augmentation transforms can now be accessed more intuitively, making it easier for developers to customize and debug pipelines.
- **Improved Pretrained Weights Loading:** Pretrained weights are now loaded directly during training, ensuring expected behavior and smoother user experience.
- **Keypoint Data Integrity:** The original keypoint data is preserved, and confidence-based filtering is handled more cleanly, improving data reliability.
- **Enhanced Keypoint Flipping:** Vertical and horizontal flip augmentations for pose estimation now work reliably, with clear warnings if required configuration is missing.
- **Smarter Predictor Handling:** YOLOE and visual prompt prediction logic is now more robust, reducing errors and improving compatibility with video and stream sources.
- **Better Validation Metrics:** Validation summaries now have clearer metric names, more detailed per-class info, and improved export options (including direct Colab links).
- **Dynamic Batch Export Fix:** Exported models now correctly handle dynamic batch sizes, preventing shape mismatches during inference.
- **Parallel Training Compatibility:** Text embedding generation is now compatible with multi-GPU setups, improving YOLO World and YOLOE training stability.
- **Consistent Object Counting Results:** Object counting results and documentation are now more consistent and easier to understand.
- **Streamlined XML Export:** The XML export process is simplified, removing unnecessary dependencies and ensuring more reliable output.
- **Documentation Improvements:** Expanded tips and examples for handling images with extreme aspect ratios in classification tasks.

🎯 Purpose & Impact
- **More Reliable Keypoint Detection:** By keeping keypoints within image bounds and preserving original data, users get higher-quality training and more accurate predictions, especially for pose estimation tasks.
- **Easier Customization:** Developers can now more easily access and modify data augmentations, leading to faster experimentation and fewer bugs.
- **Smoother Training Experience:** Loading pretrained weights and handling parallel training setups is now more intuitive and robust, reducing setup headaches.
- **Better Results Analysis:** Improved validation summaries and export options help both experts and newcomers quickly interpret and share model performance.
- **Cleaner, Faster Exports:** Streamlined XML export and dynamic batch fixes mean models are easier to deploy and integrate into diverse workflows.
- **Clearer Documentation:** Enhanced docs and code examples make it easier for all users to get started and avoid common pitfalls.

Overall, this update delivers a more stable, accurate, and user-friendly experience for anyone training, validating, or deploying YOLO models. 🎉

What's Changed
* Scope pretrained weights loading by Laughing-q in https://github.com/ultralytics/ultralytics/pull/21156
* Preserve original keypoint data and eliminate internal `0.5` threshold by WillieMaddox in https://github.com/ultralytics/ultralytics/pull/21165
* Support flip up/down augmentation for pose by WillieMaddox in https://github.com/ultralytics/ultralytics/pull/20932
* YOLOE: Optimize redundant predictor initialization by Y-T-G in https://github.com/ultralytics/ultralytics/pull/21198
* Update classification tip for `CustomizedValidator` by picsalex in https://github.com/ultralytics/ultralytics/pull/21196
* Align `metrics` summary method to `val` logs by RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/21203
* Fix dynamic batch inference with `nms=True` by Y-T-G in https://github.com/ultralytics/ultralytics/pull/21190
* YOLOE: Fix `generate_text_embeddings` for DDP training by Laughing-q in https://github.com/ultralytics/ultralytics/pull/21210
* Add `SolutionResults` attributes in `index.md` and `classwise_counts` typo fix by RizwanMunawar in https://github.com/ultralytics/ultralytics/pull/21216
* Remove unnecessary `lxml` dependency by Laughing-q in https://github.com/ultralytics/ultralytics/pull/21221
* `ultralytics 8.3.160` Clip `keypoints` for better visualization control by Laughing-q in https://github.com/ultralytics/ultralytics/pull/21220

New Contributors
* WillieMaddox made their first contribution in https://github.com/ultralytics/ultralytics/pull/21165

**Full Changelog**: https://github.com/ultralytics/ultralytics/compare/v8.3.159...v8.3.160