fix: slice numpy array values in custom_data per row in CSVSink

[farukalamai]

Before submitting

• Self-reviewed the code
• Updated documentation, follow Google-style
• Added docs entry for autogeneration (if new functions/classes)
• Added/updated tests
• All tests pass locally

Description

Fixes a bug in CSVSink and JSONSink where passing a numpy array as a
custom_data value wrote the entire array on every row instead of the
per-detection scalar value.

Type of Change

• 🐛 Bug fix (non-breaking change which fixes an issue)

Motivation and Context

When users pass computed per-detection values like detections.area via
custom_data, each row should receive its own scalar — not the whole array.

# Before (broken): every row got the full array
with sv.CSVSink("out.csv") as sink:
    sink.append(detections, custom_data={"area": detections.area})
# area column: [400.0, 400.0] on every row ❌

# After (fixed): each row gets its own value
# area column: 400.0, 400.0 ✅

The root cause was row.update(custom_data) inside the per-detection loop,
which blindly wrote the whole value. The fix applies the same per-index
slicing logic that detections.data already uses correctly.

Closes #1397

Changes Made

• src/supervision/detection/tools/csv_sink.py — slice numpy array values in custom_data per detection row
• src/supervision/detection/tools/json_sink.py — same fix
• tests/detection/test_csv.py — added test case for numpy array in custom_data

Testing

• I have tested this code locally
• I have added unit tests that prove my fix is effective or that my feature works
• All new and existing tests pass

Google Colab (optional)

Colab link:

Screenshots/Videos (optional)

Additional Notes

The fix is backward compatible — scalar values in custom_data (e.g.
{"frame_number": 42}) continue to work as before, written as-is on every
row.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 78%. Comparing base (72fc49f) to head (43da77a).
⚠️ Report is 1 commits behind head on develop.

Additional details and impacted files

@@           Coverage Diff           @@
##           develop   #2199   +/-   ##
=======================================
  Coverage       78%     78%           
=======================================
  Files           63      63           
  Lines         7972    7979    +7     
=======================================
+ Hits          6248    6257    +9     
+ Misses        1724    1722    -2     

🚀 New features to boost your workflow:

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

[Copilot]

Pull request overview

Fixes incorrect serialization of per-detection custom_data in CSVSink/JSONSink when users pass numpy arrays (previously the full array was written on every row), aligning output with expected “one value per detection row” behavior.

Changes:

• Update CSVSink.parse_detection_data() to slice custom_data numpy arrays per detection row.
• Update JSONSink.parse_detection_data() to slice custom_data numpy arrays per detection row.
• Add a unit test ensuring CSVSink slices numpy-array custom_data per row.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 5 comments.

File	Description
src/supervision/detection/tools/csv_sink.py	Slice numpy-array custom_data per detection row when producing CSV rows.
src/supervision/detection/tools/json_sink.py	Apply analogous per-row slicing for numpy-array custom_data when producing JSON rows.
tests/detection/test_csv.py	Add regression test covering numpy-array custom_data in CSVSink.

[Borda]

Thanks for the fix @farukalamai! I've pushed 3 follow-up commits to your branch:

1. fix: restore non-ndarray custom_data passthrough in CSVSink — the new else branch used hasattr(value, "__getitem__") to decide whether to slice custom_data values per detection. This accidentally indexed dicts by integer 0, raising KeyError: 0 for the existing "Complex Data" test case (custom_data={"metadata": {"sensor_id": 101}} etc.). Non-ndarray types should be written as-is; only np.ndarray values need per-row indexing. Changed to row[key] = value in the else branch.

2. test: add JSONSink test for numpy array custom_data slicing — symmetric test for JSONSink to match the CSV coverage you added.

3. test: add mixed-type custom_data test (ndarray + scalar together) — exercises all three dispatch paths at once: ndarray sliced per row + scalar broadcast to every row.

All pre-commit hooks pass and 15/15 tests pass.