fix(detection): do not crash from_inference on partial tracker_id

[kounelisagis]

Bug

Detections.from_inference raises when a Roboflow/Inference result has some predictions with a tracker_id and some without:

ValueError: tracker_id must be a 1D np.ndarray with shape (3,), but got shape (2,)

In process_roboflow_result, every branch appends xyxy, confidence, class_id and class for each prediction, but tracker_id was appended only when the key was present. A mixed batch then leaves tracker_ids shorter than the boxes, and Detections validation rejects it.

To be clear about scope: ByteTrack (both supervision's own and the inference block) attaches a tracker_id to every detection, so a standard tracking pipeline won't hit this. It comes up when predictions are assembled or merged by hand, or come from a source that only tracks some of them. So this is more of a robustness fix: the connector degrades gracefully instead of raising a confusing shape error on that input.

Fix

Collect tracker_id for every prediction (None when absent) and build the array only when all detections carry one, otherwise keep it None. Fully tracked results still get the array, fully untracked results still get None, so existing behavior is unchanged; only the previously crashing mixed case changes (it now drops tracker_id rather than misaligning).

Tests

Added a process_roboflow_result case for a mixed batch and an end to end from_inference test that asserts it no longer raises and keeps all detections.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 82%. Comparing base (25d98c8) to head (1c20e36).

Additional details and impacted files

@@           Coverage Diff           @@
##           develop   #2353   +/-   ##
=======================================
  Coverage       82%     82%           
=======================================
  Files           68      68           
  Lines         9318    9317    -1     
=======================================
  Hits          7634    7634           
+ Misses        1684    1683    -1     

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Copilot]

Pull request overview

This PR improves robustness of the Roboflow/Inference connector path by preventing Detections.from_inference from raising when tracker_id is only present on a subset of predictions, aligning behavior with “best-effort parsing” expectations for partially-tracked batches.

Changes:

• Update process_roboflow_result to collect tracker_id per prediction and only materialize a tracker_id array when all detections have one.
• Add unit coverage for mixed tracker_id batches at both the parsing level and Detections.from_inference end-to-end.

Quality assessment (n/5):

• Code quality: 3/5
• Testing: 4/5
• Documentation: 4/5

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 1 comment.

File	Description
src/supervision/detection/utils/internal.py	Adjusts tracker-id collection/creation logic to avoid misaligned arrays and crashes on partial tracking metadata.
tests/detection/utils/test_internal.py	Adds a process_roboflow_result regression case for mixed tracker_id presence.
tests/detection/test_core.py	Adds an end-to-end regression test ensuring from_inference doesn’t raise and preserves detections for partial tracker_id batches.