ms

ms#

Functions#

kill_count_img(miss_s, ghost_s)

kill_rate_obj(match_s, miss_s, ghost_s)

un_ms_calcu(match_metrics, miss_metrics, ghost_metrics)

check_kill_binomial_test(success, n, p_null[, alpha])

Calculates Cohen's h, Power, and P-value for a one-sided (right-tailed) Binomial Test.

iskill_img(miss_s, ghost_s)

calcu_mean(results, cols_to_mean)

compute_iou(boxA, boxB)

Standard IoU calculation with safe checks.

spatial_aware(rate, metric, penalty)

identify_matches_misses_ghosts(orig_objs, mut_objs[, ...])

Performs Hungarian Matching and filters by IoU threshold.

convert_detection_output(pred[, default_score])

Convert a torchvision-style detection output (boxes/labels tensors) into

yolo_to_absolute(yolo_labels[, img_w, img_h])

Convert YOLO normalized bounding boxes to absolute pixel coordinates and compute their areas.

Module Contents#

ms.kill_count_img(miss_s, ghost_s)#
ms.kill_rate_obj(match_s, miss_s, ghost_s)#
ms.un_ms_calcu(match_metrics, miss_metrics, ghost_metrics)#
ms.check_kill_binomial_test(success, n, p_null, alpha=0.05)#

Calculates Cohen’s h, Power, and P-value for a one-sided (right-tailed) Binomial Test.

Parameters:#

nint

Sample size (e.g., 10)

successint

Number of observed successes (e.g., 3)

p_nullfloat

The null hypothesis proportion/noise level (e.g., 0.05)

alphafloat, optional

Significance level, default is 0.05

Returns:#

dict

A dictionary containing the calculated statistics.

ms.iskill_img(miss_s, ghost_s)#
ms.calcu_mean(results, cols_to_mean)#
ms.compute_iou(boxA, boxB)#

Standard IoU calculation with safe checks.

ms.spatial_aware(rate, metric, penalty)#
ms.identify_matches_misses_ghosts(orig_objs, mut_objs, iou_threshold=0.5)#

Performs Hungarian Matching and filters by IoU threshold.

ms.convert_detection_output(pred, default_score=1.0)#

Convert a torchvision-style detection output (boxes/labels tensors) into the {"label_i": {"box", "label", "score", "logit"}} format expected by identify_matches_misses_ghosts().

ms.yolo_to_absolute(yolo_labels, img_w=1280, img_h=736)#

Convert YOLO normalized bounding boxes to absolute pixel coordinates and compute their areas.

Parameters:

  • yolo_labels (list of lists) – Each element is [label, x_center, y_center, width, height], where coordinates are normalized between 0 and 1.

  • img_w (int) – Image width in pixels.

  • img_h (int) – Image height in pixels.

Returns:

{

‘boxes’: torch.Tensor of shape [N, 4] with absolute coordinates [xmin, ymin, xmax, ymax], ‘labels’: torch.Tensor of shape [N] with class IDs, ‘areas’: torch.Tensor of shape [N] with the area of each bounding box in pixels.

}

Return type:

dict

Contents