perf(detection): count mask pixels with count_nonzero

[RubenHaisma]

Description

Mask pixel-area counting used np.sum: np.array([np.sum(m) for m in masks]) in Detections.area, and np.sum(mask, axis=(1, 2)) in the metrics helper get_mask_size_category.

For boolean masks, np.count_nonzero (with no axis) dispatches to NumPy's SIMD popcount over the raw byte buffer, whereas every axis-reduction form — np.sum(..., axis=...) and even np.count_nonzero(..., axis=...) — falls back to a slower generic reduction. So counting per-mask with np.count_nonzero is several times faster than the "obvious" vectorized sum, while producing bit-identical integer counts. (The plain masks.sum(axis=(1, 2)) vectorization, by contrast, is only ~1.1x — it's the primitive, not the loop-removal, that matters here.)

Both sites now use:

np.fromiter((np.count_nonzero(m) for m in masks), dtype=np.int64, count=len(masks))

dtype=np.int64 preserves the documented Detections.area mask-branch dtype on every platform (a bare np.array([...]) of Python ints is int32 on Windows).

Performance

~5x on 640×640 masks, density-independent:

path	N	before	after	speedup
Detections.area	100	7.8 ms	1.5 ms	5.3×
Detections.area	300	24.5 ms	4.6 ms	5.3×
get_mask_size_category	300	~24 ms	~4.4 ms	5.5×

get_mask_size_category feeds the size-bucketed F1Score / Precision / Recall / MeanAveragePrecision / MeanAverageRecall metrics, where it runs repeatedly (SMALL/MEDIUM/LARGE × predictions+targets) across a dataset.

Correctness

Counting True pixels equals summing a bool array exactly (integers, no float error). Verified bit-identical over 400 randomized trials plus empty / all-true / all-false / 1×1 edge cases, with int64 dtype preserved. Existing Detections.area and metrics suites pass unchanged.

Tests

Adds parity tests for Detections.area (dense mask) and get_mask_size_category, each against an np.sum reference, plus threshold-boundary coverage. ruff check/format clean.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 84%. Comparing base (8f576b0) to head (d42aead).

Additional details and impacted files

@@           Coverage Diff           @@
##           develop   #2361   +/-   ##
=======================================
  Coverage       84%     84%           
=======================================
  Files           70      70           
  Lines         9923    9926    +3     
=======================================
+ Hits          8313    8316    +3     
  Misses        1610    1610           

🚀 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 mask pixel-area counting performance in Supervision’s detection and metrics code paths by switching from np.sum reductions to a faster per-mask np.count_nonzero + np.fromiter approach, while preserving bit-identical results (for boolean masks) and maintaining an int64 area dtype.

Changes:

• Updated Detections.area (mask branch) to compute per-mask areas via np.fromiter((np.count_nonzero(m) ...), dtype=np.int64).
• Updated get_mask_size_category to use the same count_nonzero + fromiter strategy for faster size bucketing.
• Added tests to verify parity with np.sum references and to cover threshold boundary behavior and dense-mask area expectations.

Reviewed changes

Copilot reviewed 4 out of 5 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
src/supervision/detection/core.py	Speeds up Detections.area for boolean masks using count_nonzero + fromiter with stable int64 output.
src/supervision/metrics/utils/object_size.py	Speeds up mask-size categorization by computing areas via per-mask count_nonzero + fromiter.
tests/detection/test_core.py	Adds a dense-mask parity test to ensure Detections.area matches pixel counts / sum and remains int64.
tests/metrics/utils/test_object_size.py	Adds reference-parity and boundary tests for get_mask_size_category.
tests/metrics/utils/__init__.py	Adds/keeps the tests.metrics.utils package marker (empty file).