fix(detection): preserve class_name string dtype on empty Detections.from_inference

[Ace3Z]

For non-empty inference results, Detections.from_inference stores class_name as a string array (e.g. <U3 for ["cat"]). On the empty short-circuit path it instead built {CLASS_NAME_DATA_FIELD: np.empty(0)}, which defaults to float64. That breaks downstream code that assumes the array is string-typed (e.g. np.isin(det["class_name"], ["cat"]) returns a mismatched dtype warning, indexing into a list of names crashes, etc).

process_roboflow_result already returns a correctly typed empty data dict ({CLASS_NAME_DATA_FIELD: np.empty(0, dtype=str)} on the empty short-circuit, line 71 of detection/utils/internal.py). The fix is to use that returned data directly instead of rebuilding a wrong-dtype dict.

Minimal repro:

import supervision as sv

empty = sv.Detections.from_inference({"predictions": [], "image": {"width": 100, "height": 100}})
non_empty = sv.Detections.from_inference({
    "predictions": [{"x": 50, "y": 50, "width": 20, "height": 20,
                     "confidence": 0.9, "class": "cat", "class_id": 0}],
    "image": {"width": 100, "height": 100},
})
print(empty["class_name"].dtype, non_empty["class_name"].dtype)
# before: float64 <U3
# after:  <U1 <U3

tests/detection/test_core.py::test_from_inference_empty_class_name_dtype_matches_non_empty asserts both paths produce string-kind arrays.

[Copilot]

Pull request overview

Fixes a dtype inconsistency in Detections.from_inference so that the class_name data field remains string-typed even when the Roboflow/Inference result contains zero predictions, preventing downstream dtype-mismatch warnings/errors.

Changes:

• Reuse process_roboflow_result’s returned data dict in the empty short-circuit path instead of rebuilding it with np.empty(0) (which defaults to float64).
• Add a regression test ensuring empty and non-empty inference results produce class_name arrays of the same string dtype kind.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
src/supervision/detection/core.py	Uses process_roboflow_result’s data directly for empty inference results to preserve string dtype.
tests/detection/test_core.py	Adds coverage asserting class_name dtype kind matches between empty and non-empty inference outputs.

[codecov]

Codecov Report

❌ Patch coverage is 14.28571% with 6 lines in your changes missing coverage. Please review.
✅ Project coverage is 78%. Comparing base (2fdb970) to head (2c8a7ab).

Additional details and impacted files

@@           Coverage Diff           @@
##           develop   #2270   +/-   ##
=======================================
  Coverage       78%     78%           
=======================================
  Files           66      66           
  Lines         8412    8416    +4     
=======================================
+ Hits          6580    6589    +9     
+ Misses        1832    1827    -5     

🚀 New features to boost your workflow:

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