记录手撕iou面试代码

在这个函数中，boxA 和 boxB 是包含四个整数的列表，代表边界框的四个坐标： [开始x, 开始y, 结束x, 结束y]。

这个函数首先计算两个边界框的交集面积，然后计算每个边界框的面积和两个边界框的并集面积，最后用交集面积除以并集面积，得到的结果就是IOU。

def bbox_intersection_over_union(boxA, boxB):
    # 确定每个边界框的 (y, x) 的 (开始，结束) 坐标
    xA = max(boxA[0], boxB[0])
    yA = max(boxA[1], boxB[1])
    xB = min(boxA[2], boxB[2])
    yB = min(boxA[3], boxB[3])

    # 计算交集的面积
    interArea = max(0, xB - xA + 1) * max(0, yB - yA + 1)

    # 计算每个边界框的面积
    boxAArea = (boxA[2] - boxA[0] + 1) * (boxA[3] - boxA[1] + 1)
    boxBArea = (boxB[2] - boxB[0] + 1) * (boxB[3] - boxB[1] + 1)

    # 计算并集的面积
    unionArea = boxAArea + boxBArea - interArea

    # 计算交并比
    iou = interArea / float(unionArea)

    # 返回交并比
    return iou

• xA 是两个边界框左上角 x 坐标的较大值，表示交集的左上角 x 坐标。
• yA 是两个边界框左上角 y 坐标的较大值，表示交集的左上角 y 坐标。
• xB 是两个边界框右下角 x 坐标的较小值，表示交集的右下角 x 坐标。
• yB 是两个边界框右下角 y 坐标的较小值，表示交集的右下角 y 坐标。

这些计算是为了找到两个边界框的交集部分的坐标范围，这将用于计算交集的面积。如果 xA 大于 xB 或者 yA 大于 yB，说明两个边界框没有交集，此时交集的面积将为0。

参考其他大佬的：

# -*- coding: utf-8 -*-
#
# This is the python code for calculating bbox IoU,
# By running the script, we can get the IoU score between pred / gt bboxes
#
# Author: hzhumeng01 2018-10-19
# copyright @ netease, AI group

from __future__ import print_function, absolute_import
import numpy as np

def get_IoU(pred_bbox, gt_bbox):
    """
    return iou score between pred / gt bboxes
    :param pred_bbox: predict bbox coordinate
    :param gt_bbox: ground truth bbox coordinate
    :return: iou score
    """

    # bbox should be valid, actually we should add more judgements, just ignore here...
    # assert ((abs(pred_bbox[2] - pred_bbox[0]) > 0) and
    #         (abs(pred_bbox[3] - pred_bbox[1]) > 0))
    # assert ((abs(gt_bbox[2] - gt_bbox[0]) > 0) and
    #         (abs(gt_bbox[3] - gt_bbox[1]) > 0))

    # -----0---- get coordinates of inters
    ixmin = max(pred_bbox[0], gt_bbox[0])
    iymin = max(pred_bbox[1], gt_bbox[1])
    ixmax = min(pred_bbox[2], gt_bbox[2])
    iymax = min(pred_bbox[3], gt_bbox[3])
    iw = np.maximum(ixmax - ixmin + 1., 0.)
    ih = np.maximum(iymax - iymin + 1., 0.)

    # -----1----- intersection
    inters = iw * ih

    # -----2----- union, uni = S1 + S2 - inters
    uni = ((pred_bbox[2] - pred_bbox[0] + 1.) * (pred_bbox[3] - pred_bbox[1] + 1.) +
           (gt_bbox[2] - gt_bbox[0] + 1.) * (gt_bbox[3] - gt_bbox[1] + 1.) -
           inters)

    # -----3----- iou
    overlaps = inters / uni

    return overlaps

def get_max_IoU(pred_bboxes, gt_bbox):
    """
    given 1 gt bbox, >1 pred bboxes, return max iou score for the given gt bbox and pred_bboxes
    :param pred_bbox: predict bboxes coordinates, we need to find the max iou score with gt bbox for these pred bboxes
    :param gt_bbox: ground truth bbox coordinate
    :return: max iou score
    """

    # bbox should be valid, actually we should add more judgements, just ignore here...
    # assert ((abs(gt_bbox[2] - gt_bbox[0]) > 0) and
    #         (abs(gt_bbox[3] - gt_bbox[1]) > 0))

    if pred_bboxes.shape[0] > 0:
        # -----0---- get coordinates of inters, but with multiple predict bboxes
        ixmin = np.maximum(pred_bboxes[:, 0], gt_bbox[0])
        iymin = np.maximum(pred_bboxes[:, 1], gt_bbox[1])
        ixmax = np.minimum(pred_bboxes[:, 2], gt_bbox[2])
        iymax = np.minimum(pred_bboxes[:, 3], gt_bbox[3])
        iw = np.maximum(ixmax - ixmin + 1., 0.)
        ih = np.maximum(iymax - iymin + 1., 0.)

        # -----1----- intersection
        inters = iw * ih

        # -----2----- union, uni = S1 + S2 - inters
        uni = ((gt_bbox[2] - gt_bbox[0] + 1.) * (gt_bbox[3] - gt_bbox[1] + 1.) +
               (pred_bboxes[:, 2] - pred_bboxes[:, 0] + 1.) * (pred_bboxes[:, 3] - pred_bboxes[:, 1] + 1.) -
               inters)

        # -----3----- iou, get max score and max iou index
        overlaps = inters / uni
        ovmax = np.max(overlaps)
        jmax = np.argmax(overlaps)

    return overlaps, ovmax, jmax

if __name__ == "__main__":

    # test1
    pred_bbox = np.array([50, 50, 90, 100])   # top-left: <50, 50>, bottom-down: <90, 100>, <x-axis, y-axis>
    gt_bbox = np.array([70, 80, 120, 150])

    print (get_IoU(pred_bbox, gt_bbox))

    # test2
    pred_bboxes = np.array([[15, 18, 47, 60],
                          [50, 50, 90, 100],
                          [70, 80, 120, 145],
                          [130, 160, 250, 280],
                          [25.6, 66.1, 113.3, 147.8]])
    gt_bbox = np.array([70, 80, 120, 150])
    print (get_max_IoU(pred_bboxes, gt_bbox))