Changelog¶
All notable changes to YOLOmatic are documented here. The format follows Keep a Changelog, and this project adheres to Semantic Versioning.
[Unreleased] - 2026-06-10¶
Added¶
- Added split-aware Ultralytics NDJSON conversion to YOLO Pose and COCO Pose, including keypoint metadata and matching TUI/Platform format choices.
- Added support for YOLO pose estimation tasks (YOLO26-pose, YOLOv11-pose, YOLOv8-pose model families).
- Implemented authoritative pose detection via
kpt_shapein dataset yaml. - Added TensorRT dynamic batch compilation mode (
trt_dynamic_batchparameter). - Added caching mechanism for dataset summaries to reduce redundant computations and disk I/O.
Changed¶
- Updated
wraptdependency to 2.2.0 inuv.lock. - Preserved user-selected TensorRT export workspace values instead of clamping them to a local GPU memory heuristic.
- Made the standalone export wizard load checkpoint details before export and hide options that do not apply to the selected model task or target format.
Fixed¶
- Serialized Windows OpenCV initialization before Albumentations augmentation workers and standardized safe TUI back/cancel shortcuts.
- Ensured live TUI updates refresh the UI immediately to prevent stale rendering.
[v5.0.0] - 2026-05-27¶
Added¶
- Implemented interactive dataset conversion wizard for NDJSON formats.
- Added support for Ultralytics-platform NDJSON conversion and header parsing.
- Implemented rare-class seeding for smart balanced dataset splits.
- Added flat-structure fallback for datasets lacking explicit split keys.
- Added navigation support to CLI configuration wizards using step-machine logic.
- Added progress reporting to dataset splitting functions.
- Added multiprocessing support for NDJSON dataset preparation.
Changed¶
- Refactored CLI to simplify rich text handling.
- Optimized TUI rendering and layout performance with layout caching.
- Parallelized dataset inspection and directory listing using
ThreadPoolExecutor. - Standardized YOLO label directory resolution and parsing logic.
- Updated project branding assets, including favicons and logo formats.
- Restructured documentation into an MkDocs Material site.
- Updated dependencies including
tensorflow,softprops/action-gh-release,actions/checkout,actions/setup-python, andastral-sh/setup-uv.
Fixed¶
- Guarded conversion result elapsed display in CLI.
- Corrected demo GIF filenames and image links in README.
- Removed unused imports and constants across core and CLI modules.
- Disabled MLflow callback in Ultralytics runtime to prevent import conflicts.
[v4.4.0] - 2026-05-20¶
Added¶
- Implemented Labelbox NDJSON to YOLO/COCO dataset converter.
- Added unified augmentation profile editor with integrated guidance.
- Added SAM 3.1 support, benchmarking, and augmentation capabilities.
- Implemented offline Albumentations-based dataset augmentation engine.
- Added max workers option to benchmark CLI for parallel execution.
- Added prediction confidence visualization to benchmark reports.
Changed¶
- Upgraded project to Python 3.12.
- Overhauled report UI grouping and TUI layout proportions.
- Replaced object size sensitivity heatmap with grouped horizontal bar charts.
- Updated dependencies including
blessed,importlib-metadata, andultralytics.
Fixed¶
- Updated benchmark tests for Plotly JSON encoding compatibility.
- Cleared environment variables when mocking subprocess in update flow tests.
- Removed unused variables and imports in augmentation modules.
[v4.3.0] - 2026-05-14¶
Added¶
- Implemented benchmark model evaluation and vector analysis feature.
- Added SAM 3.1 segmentation inference and fine-tuning commands.
- Added "Finish" option to TUI menus for streamlined workflows.
- Implemented dynamic color interpolation for metrics in HTML reports.
- Added full-screen transient live progress display for CLI benchmarks.
Changed¶
- Refined benchmark table layout and dynamic sizing.
- Migrated dependency upgrade logic to
uv. - Updated Python requirement to 3.12.
Fixed¶
- Resolved circular dependencies in benchmark and CLI modules.
- Removed unused imports and variables across core modules.
[v4.2.0 and earlier]¶
Earlier history is captured in git log. The first public release added Labelbox NDJSON → YOLO/COCO conversion, multi-model config generation, ClearML integration, and TensorBoard launcher.