Prebuilt package version 1.2.0.
- Modernized the project structure and developer workflow with
uv, a Makefile, Ruff-based formatting/linting, expanded tests, release validation, and generated documentation tooling - Added official Python 3.12 support and reworked optional dependency groups so features such as RealSense, OpenVINO, MediaPipe, DepthAI, and other integrations can be installed independently
- Expanded the input system with a unified depth-camera example and made OpenCV video-capture backend selection platform-aware, exposing only backends supported by the installed OpenCV build
- Reworked the public API and examples toward direct context-managed inputs and estimators while retaining composable graph-based processing for more complex pipelines
- Improved model asset handling with on-demand downloads, a configurable asset directory, and centralized Hugging Face-hosted model storage
- Added comprehensive model, third-party code, image, and documentation asset attribution, including explicit licensing information for bundled tracking utilities and externally downloaded models
- Expanded documentation with clearer installation, input, development, build, testing, model-license, and example guidance
- Updated and reorganized examples for object detection, tracking, pose estimation, depth cameras, segmentation, inpainting, monocular depth, RGB-D processing, and related real-time computer-vision workflows