COCOeval

class faster_coco_eval.core.cocoeval.COCOeval(cocoGt=None, cocoDt=None, iouType='segm', ranges=None, print_function=<bound method Logger.info of <Logger faster_coco_eval.core.cocoeval (WARNING)>>, extra_calc=False, kpt_oks_sigmas=None, use_area=True, lvis_style=False, separate_eval=False, boundary_dilation_ratio=0.02, boundary_cpu_count=4, rle_iou_max_workers=8)

Bases: object

accumulate(p=None)

Deprecated. Use COCOeval_faster.accumulate instead.

Parameters:

p – Unused.

Raises:

DeprecationWarning – Always.

computeIoU(imgId, catId)

Compute IoU between ground truth and detection for a given image and category.

Parameters:

  • imgId (int) – Image ID.

  • catId (int) – Category ID.

Returns:

IoUs between gt and dt for the given image and category.

Return type:

list[float] | ndarray

computeOks(imgId, catId)

Compute OKS between ground truth and detection for a given image and category.

Parameters:

  • imgId (int) – Image ID.

  • catId (int) – Category ID.

Returns:

OKS between gt and dt for the given image and category.

Return type:

ndarray

evaluate()

Deprecated. Use COCOeval_faster.evaluate instead.

Raises:

DeprecationWarning – Always.

evaluateImg(imgId, catId, aRng, maxDet)

Deprecated. Use COCOeval_faster.evaluateImg instead.

Parameters:

  • imgId – Image ID.

  • catId – Category ID.

  • aRng – Area range.

  • maxDet – Maximum detections.

Raises:

DeprecationWarning – Always.

get_type_result(first=0.01, second=0.85)

Calculate type results for easy, medium, and hard splits.

Parameters:

  • first (float) – Threshold for ‘easy’ crowdIndex.

  • second (float) – Threshold for ‘medium’ crowdIndex.

Returns:

List of scores for [easy, medium, hard].

Return type:

list

property print_function: Callable

Returns: Callable: The function used for printing/logging.

split(gt_file, first=0.01, second=0.85)

Split images into easy, medium, hard according to ‘crowdIndex’.

Parameters:

  • gt_file (str) – Path to the ground truth file.

  • first (float) – Threshold for ‘easy’ crowdIndex.

  • second (float) – Threshold for ‘medium’ crowdIndex.

Returns:

Lists of image ids for (easy, medium, hard).

Return type:

tuple

summarize()

Compute and display summary metrics for evaluation results. After calling this method, self.all_stats will contain the full results including all metrics while self.stats contains a subset of the most commonly used metrics.

The summary layout adapts to the configured area ranges and maximum detection counts. stats contains the metrics before the IoU-specific recall summaries, while all_stats contains every generated metric.