- Updated type hints across multiple files to replace `Optional[X]` with `X | None` for better readability and consistency.
- Added `from __future__ import annotations` to files to enable postponed evaluation of type annotations.
- Adjusted function signatures and variable types in various modules including `polygon_zone.py`, `smoother.py`, `transformers.py`, `vlm.py`, and others.
- Ensured compatibility with Python 3.10+ type hinting features.
Signed-off-by: Onuralp SEZER <thunderbirdtr@gmail.com>
- separated args list in examples between inference and ultralytics
- added depricated decorator to from_roboflow
- use `with supress(...)` for parsing inference response models instead of checking instance type (no more pydantic dep)
- a few other README changes
Restructuring the video stream and traffics analysis codes resulted in two new key classes, VideoProcessor, and DetectionsManager. The main aim behind this was to improve code readability and reusability. VideoProcessor class handles video processing as an object, while DetectionsManager is responsible for managing detections and trackings. This change has also led to the removal of the process_frame function, as its functionality is now encapsulated within VideoProcessor class methods. Moreover, the code now identifies zones in and out separately, thereby handling detections more distinctly and efficiently. Lastly, Polygon Zones are no longer differentiated by color.
The change modifies the polygon zone verification method in `polygon_zone.py` to ensure `self.current_count` always returns an integer, preventing possible type errors in downstream processing. Two classes, `Trace` and `TraceAnnotator`, were added to `annotate.py`, providing functionality for trace annotation during detection. In `script.py`, optimized the trace annotation process by replacing the previous class implementations with the new ones from `annotate.py`, and also cleaned up and simplified polygon zone initiation and frame processing. This provides more efficient and robust video processing and zone-based detection features."
Added two new classes to enhance traffic analysis over video frames. Trace class stores the positional information of different objects detected in each video frame for a specified length. TraceAnnotator class uses the information stored in Trace instance to annotate the motion path of detected objects on the video frames. Also, corrected the incorrect numpy import statement in the dependence code.
Onuralp SEZER
LinasKo
SkalskiP
pre-commit-ci[bot]
Jeslin P James
Kokouvi
bright1205
Paul
Hugo Dutra