COCOeval_faster

class faster_coco_eval.core.COCOeval_faster(cocoGt=None, cocoDt=None, iouType='segm', 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)

Bases: COCOeval

This is a slightly modified version of the original COCO API, where the functions evaluateImg() and accumulate() are implemented in C++ to speedup evaluation.

accumulate()

Accumulate per image evaluation results and store the result in self.eval.

Does not support changing parameter settings from those used by self.evaluate()

static calc_auc(recall_list, precision_list, method='c++')

Calculate area under precision recall curve.

Parameters:

  • recall_list (Union[List[float], np.ndarray]) – List or array of recall values.

  • precision_list (Union[List[float], np.ndarray]) – List or array of precision values.

  • method (str, optional) – Method to calculate auc. Defaults to “c++”.

Returns:

Area under precision recall curve.

Return type:

float

compute_mAUC()

Compute the mean Area Under Curve (mAUC) metric.

Returns:

Mean AUC across all categories and area ranges.

Return type:

float

compute_mIoU()

Compute the mean Intersection over Union (mIoU) metric.

Returns:

Mean IoU across all matched detections and ground truths.

Return type:

float

evaluate()

Run per image evaluation on given images and store results in self.evalImgs_cpp, a datastructure that isn’t readable from Python but is used by a c++ implementation of accumulate().

Unlike the original COCO PythonAPI, we don’t populate the datastructure self.evalImgs because this datastructure is a computational bottleneck.

property extended_metrics

Computes extended evaluation metrics for object detection results.

Calculates per-class and overall (macro) metrics such as mean average precision (mAP) at IoU thresholds, precision, recall, and F1-score. Results are computed using evaluation results stored in the object. For each class, if categories are used, metrics are reported separately and for the overall dataset.

Returns:

A dictionary with the following keys:
  • ’class_map’ (list of dict): List of per-class and overall metrics, each as a dictionary containing:

    • ’class’ (str): Class name or “all” for macro metrics.

    • ’map@50:95’ (float): Mean average precision at IoU 0.50:0.95.

    • ’map@50’ (float): Mean average precision at IoU 0.50.

    • ’precision’ (float): Macro-averaged precision.

    • ’recall’ (float): Macro-averaged recall.

  • ’map’ (float): Overall mean average precision at IoU 0.50.

  • ’precision’ (float): Macro-averaged precision for the best F1-score.

  • ’recall’ (float): Macro-averaged recall for the best F1-score.

Return type:

dict

Notes

  • Uses COCO-style evaluation results (precision and scores arrays).

  • Filters out classes with NaN results in any metric.

  • The best F1-score across recall thresholds is used to select macro precision and recall.

math_matches()

Analyze matched detections and ground truths to assign true positive, false positive, and false negative flags, and update detection and ground truth annotations in-place.

Returns:

None

run()

Wrapper function which runs the evaluation.

Calls evaluate(), accumulate(), and summarize() in sequence.

Returns:

None

property stats_as_dict

Return the evaluation statistics as a dictionary with descriptive labels.

Returns:

Dictionary mapping metric names to their values.

Return type:

dict[str, float]

summarize()

Summarize and finalize the statistics of the evaluation.

Returns:

None