Verify that get_polygon_center correctly calculates the centroid of a polygon. Scenario: Calculating the center point (centroid) of various

Verify that get_video_frames_generator yields frames with correct shapes. Scenario: Iterating over video frames using a generator. Expec

Verify that get_video_frames_generator respects start and end frame indices. Scenario: Iterating over a specific range of video frames using

Verify that get_video_frames_generator correctly handles the stride parameter. Scenario: Iterating over video frames with specified stride (

Ensures that `Detections.__getitem__` (indexing/slicing) works correctly for various input types. This is a core feature that allows users to

Two near-identical OBBs should be merged into one group; an X-pattern pair should produce two separate groups.

Two unit-height 2x2 squares offset by 1 share half their area: IoU=1/3.

Multiple OBBs in one `Detections` each get their own correct area. Guards against the shoelace reduction collapsing across boxes instead

Calling _get_logger twice with the same name does not add extra handlers.

Identical/all-True → 1.0; all-False/disjoint → 0.0; empty keeps shape.

IoU of a mask with itself must be 1.0.

Identity resize (same dimensions) preserves the decoded mask.

Resize to same shape returns identical RLE, offsets, and crop shapes.

Create a simple blank test image fixture

Test that ImageAssets.list() returns all image filenames.

Test that ImageAssets enum members have correct attributes.

cm[i] must equal masks[i] for every index.

Regression for issue #1679: the SAHI workflow with OBB detections dropped valid detections at the merge step because `with_nms`/`with_nmm` his

Test that F1Score can be initialized with custom parameters

Test that Recall can be initialized with custom parameters

Test that Precision can be initialized with custom parameters

Test that F1Score can be initialized with default parameters

Test that Recall can be initialized with default parameters

Test that Precision can be initialized with default parameters

Test download_assets with invalid asset name.

Test download_assets with invalid asset name that already exists.

Resize with non-positive dimensions raises ValueError.

BlurAnnotator must reject kernel_size < 1 at construction time.

2D and 4D inputs must raise ValueError before reaching shape[1:] check.

PixelateAnnotator must reject pixel_size < 1 at construction time.

Inputs with wrong shape raise ValueError before any computation.

IOS = intersection / min(area_a, area_b) must match dense reference.

is_compressed_rle returns True for str/bytes, False otherwise.

Verify is_empty() returns True iff the Detections object has zero detections.

IoU is invariant under uniform scaling and translation. Both are affine maps, so exact polygon IoU matches the small-coordinate basel

_rle_join_cols merges True/True boundary; no zero-length False run inserted. When column A ends True and column B starts True (leading False

_json_default raises TypeError for non-numpy objects.

NumPy array nested inside a custom_data dict value serializes as a JSON array.

NumPy scalar in custom_data serializes as a JSON number or boolean.

A single OBB prediction is always kept, regardless of threshold.

X-pattern: two thin rectangles crossing at +/-45° share an AABB but barely overlap as OBBs. AABB-NMS would suppress one; OBB-NMS must keep

Test the as_detections method for KeyPoints objects.

Confidence mean uses only the selected keypoint columns, not all.

Test the as_detections method for empty KeyPoints objects.

A [0, 0] keypoint is treated as missing and excluded from the box.

Batch with one all-zero skeleton: invalid skeleton gets box zeroed.

Only the selected keypoints contribute to the bounding box.

detection_confidence is preferred over the keypoint-confidence mean.

Test the as_detections method preserves data.

Test the creation and behavior of an empty KeyPoints object.

Test the equality comparison for KeyPoints objects.

Test the equality comparison for KeyPoints objects with data.

Detection-level filtering mirrors Detections API patterns.

Test the is_empty method for KeyPoints objects.

Test the iteration over KeyPoints objects.

Test the iteration over KeyPoints objects without confidence.

Test the __setitem__ method for KeyPoints objects.

Custom __init__ must assign every dataclass field — guards against drift.

1x1 source resized to large shape fills entirely True or False.

Mismatched predictions and oriented_boxes must fail loudly, not silently misalign rows.

RGBA input: padded alpha=0, interior alpha preserved, input array not mutated.

Extension matching normalizes leading dots, case, and multi-part extensions.

Integer list indexing must match dense fancy indexing.

Test F1 score with list inputs

Test recall with list inputs

Test precision with list inputs

Accept a legitimate nested file_name inside images/ without raising.