Update bbox chunks recovery in VideoMatting

[demoulinv]

This pull request introduces improvements to how segmentation prompts are handled and optimizes the extraction of tracking data by supporting merged tracking directions. The main changes focus on enhancing metadata generation and simplifying the logic for extracting object tracking information.

Metadata and prompt handling:

• In VideoMatting.py, the segmentation prompt is now extracted from bounding box keys and stored in the metadata_deep_model_base under Meshroom:mrSegmentation:Prompt, ensuring that prompt information is consistently recorded.

Tracking extraction logic:

• In bboxUtils.py, the extract_tracking function now supports a merged direction in the tracking data. If present, it uses merged tracking data instead of separate forward/backward directions, simplifying the merging logic and making the code more robust for different tracking data formats.
• The logic for collecting object IDs and frames is streamlined: when merged data exists, only those keys are used; otherwise, the function falls back to the previous forward direction logic.

[Copilot]

Pull request overview

This PR updates how segmentation prompt metadata is recorded in the VideoMatting pipeline and adjusts bounding-box tracking extraction to support a merged tracking direction format.

Changes:

• Record segmentation prompts derived from bbox keys into metadata_deep_model_base["Meshroom:mrSegmentation:Prompt"] in VideoMatting.py.
• Update extract_tracking in bboxUtils.py to look for merged tracking data and simplify object/frame collection logic.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 1 comment.

File	Description
segmentationRDS/bboxUtils.py	Changes tracking extraction to prefer merged direction data when present.
meshroom/videoMatting/VideoMatting.py	Extracts prompt names from bbox keys and stores them in output metadata.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.