Evaluation Metrics

fgvc.core.metrics.binary_segmentation_scores(preds: ndarray, targs: ndarray, reduction: str = 'mean') → dict

Compute segmentation scores Balanced Accuracy, Precision, Recall, F1, and IoU.

Parameters:

• preds – Array with predicted values.

• targs – Array with target values.

• reduction – Whether to aggregate scores using mean, sum, or not aggregate at all.

Returns:

A dictionary with segmentation scores.

Return type:

scores

fgvc.core.metrics.binary_segmentation_tp_fp_fn_tn(preds: ndarray, targs: ndarray, pos_label: int = 1) → Tuple[ndarray, ndarray, ndarray, ndarray]

Compute values of confusion matrix for binary segmentation.

Parameters:

• preds – Array [b, (2), h, w] with predicted values.

• targs – Array [b, h, w] with target values.

• pos_label – Class to report in the binary classification.

Returns:

• tp – Array [b,] with True Positives for each sample.

• fp – Array [b,] with False Positives for each sample.

• fn – Array [b,] with False Negatives for each sample.

• tn – Array [b,] with True Negatives for each sample.

fgvc.core.metrics.classification_scores(preds: ndarray, targs: ndarray, *, top_k: int | None = 3, return_dict: bool = True) → dict | Tuple[float, float, float]

Compute top-1 and top-k accuracy and f1 score.

Parameters:

• preds – Numpy array with predictions.

• targs – Numpy array with ground-truth targets.

• top_k – Value of k to compute top k accuracy.

• return_dict – If True, the method returns dictionary with metrics.

Returns:

A dictionary or tuple with classification scores.

Return type:

scores