Binary Class Segmentation Metrics(f1, acc, precision, recall) in Numpy

binSegMetrics.py

# https://github.com/Abe404/segmentation_of_roots_in_soil_with_unet/blob/master/src/metrics.py

def get_metrics(y_pred, y_true, loss=float('nan')):
    true_positives = np.sum(np.logical_and(y_pred == 1, y_true == 1))
    true_negatives = np.sum(np.logical_and(y_pred == 0, y_true == 0))
    false_positives = np.sum(np.logical_and(y_pred == 1, y_true == 0))
    false_negatives = np.sum(np.logical_and(y_pred == 0, y_true == 1))
    accuracy = (true_positives + true_negatives) / len(y_true)
    assert not np.isnan(true_negatives)
    assert not np.isnan(false_positives)
    assert not np.isnan(false_negatives)
    assert not np.isnan(true_positives)

    if true_positives > 0:
        precision = true_positives / (true_positives + false_positives)
        recall = true_positives / (true_positives + false_negatives)
        f1_score = 2 * ((precision * recall) / (precision + recall))
        iou = true_positives / (true_positives + false_positives + false_negatives)
    else:
        precision = recall = f1_score = iou = float('NaN')
    return {
        "accuracy": accuracy,
        "TN": true_negatives,
        "FP": false_positives,
        "FN": false_negatives,
        "TP": true_positives,
        "precision": precision,
        "recall": recall,
        "f1_score": f1_score,
        "iou": iou,
        "true_mean": np.mean(y_true),
        "pred_mean": np.mean(y_pred),
        "loss": loss
    }