The calculation of Average Precision is wrong.

[Uchan1996]

The calculation of Average Precision in compute_ap is wrong.
The current calculation outputs small AP since the integral range is narrower than [0, 1].
I think the correct calculation is as below.
Please let me know if my calculation is wrong.

def compute_ap(pred_annos, num_gt):
    pred_annos = sorted(pred_annos, key=cmp_to_key(cmp))

    num_tp = np.zeros(len(pred_annos))
    for i in range(len(pred_annos)):
        num_tp[i] = 0 if i == 0 else num_tp[i - 1]
        if pred_annos[i]["type"] == "tp":
            num_tp[i] += 1
    # logger.debug("num tp = {}".format(num_tp))

    precision = num_tp / np.arange(1, len(pred_annos) + 1) # Precision = TP / (TP + FP)
    recall = num_tp / num_gt # Recall = TP / (TP + FN)

    recall = np.concatenate([[0], recall, [1]])
    precision = np.concatenate([precision, [precision[-1]]])

    for i in range(len(pred_annos) - 1, 0, -1):
        precision[i - 1] = max(precision[i], precision[i - 1])

    dr = recall[1:] - recall[:-1]
    ap = np.sum(precision * dr)
    return ap