f1，精度，召回对象检测

Question

我正在尝试为我的物体检测模型计算精度，召回率，f1分数。我想知道我的方法是否正确吗？

def f1_eval(bbox_pred, bbox_gt, cls_pred, cls_gt, metric_iou=0.5):
    """

    :param bbox_pred: [M x4]
    :param bbox_gt: [N x 4]
    :param cls_pred: [M]
    :param cls_gt: [N]
    :return:
    """

    overlap = box_overlap(bbox_pred, bbox_gt)
    max_overlap = np.argmax(overlap, 1)
    num_of_preds = bbox_pred.shape[0]
    num_of_gts = bbox_gt.shape[0]
    TP = 0
    FP = 0
    FN = 0
    for i in range(num_of_preds):
        max_overlap_value = overlap[i, max_overlap[i]]
        if max_overlap_value >= metric_iou:
            if cls_pred[i] == cls_gt[max_overlap[i]]:
                TP +=1
            else:
                FN +=1
        else:
            FP +=1
    recall = TP / (TP + FN)
    precision = TP / (TP + FP)


    f1 = 2 * recall * precision / (recall + precision)


    return f1, precision, recall