本文系转载,出处:http://www.manongjc.com/article/37951.html
交并比(Intersection-over-Union,IoU),目标检测中使用的一个概念,是产生的候选框(candidate bound)与原标记框(ground truth bound)的交叠率,即它们的交集与并集的比值。最理想情况是完全重叠,即比值为1。
计算公式:
代码实现1:
def calculateIoU(candidateBound, groundTruthBound):
cx1 = candidateBound[0]
cy1 = candidateBound[1]
cx2 = candidateBound[2]
cy2 = candidateBound[3]
gx1 = groundTruthBound[0]
gy1 = groundTruthBound[1]
gx2 = groundTruthBound[2]
gy2 = groundTruthBound[3]
carea = (cx2 - cx1) * (cy2 - cy1) #C的面积
garea = (gx2 - gx1) * (gy2 - gy1) #G的面积
x1 = max(cx1, gx1)
y1 = max(cy1, gy1)
x2 = min(cx2, gx2)
y2 = min(cy2, gy2)
w = max(0, x2 - x1)
h = max(0, y2 - y1)
area = w * h #C∩G的面积
iou = area / (carea + garea - area)
return iou
代码实现2:
import numpy as np
def compute_iou(box1, box2, wh=False):
"""
compute the iou of two boxes.
Args:
box1, box2: [xmin, ymin, xmax, ymax] (wh=False) or [xcenter, ycenter, w, h] (wh=True)
wh: the format of coordinate.
Return:
iou: iou of box1 and box2.
"""
if wh == False:
xmin1, ymin1, xmax1, ymax1 = box1
xmin2, ymin2, xmax2, ymax2 = box2
else:
xmin1, ymin1 = int(box1[0]-box1[2]/2.0), int(box1[1]-box1[3]/2.0)
xmax1, ymax1 = int(box1[0]+box1[2]/2.0), int(box1[1]+box1[3]/2.0)
xmin2, ymin2 = int(box2[0]-box2[2]/2.0), int(box2[1]-box2[3]/2.0)
xmax2, ymax2 = int(box2[0]+box2[2]/2.0), int(box2[1]+box2[3]/2.0)
## 获取矩形框交集对应的左上角和右下角的坐标(intersection)
xx1 = np.max([xmin1, xmin2])
yy1 = np.max([ymin1, ymin2])
xx2 = np.min([xmax1, xmax2])
yy2 = np.min([ymax1, ymax2])
## 计算两个矩形框面积
area1 = (xmax1-xmin1) * (ymax1-ymin1)
area2 = (xmax2-xmin2) * (ymax2-ymin2)
inter_area = (np.max([0, xx2-xx1])) * (np.max([0, yy2-yy1])) #计算交集面积
iou = inter_area / (area1+area2-inter_area+1e-6) #计算交并比
return iou
在目标检测中,常会利用非极大值抑制算法(NMS)对生成的大量候选框进行后处理,去除冗余的候选框,得到最具代表性的结果,以加快目标检测的效率。NMS其实是为了消除多余的候选框,找到最佳的bbox。
NMS的算法逻辑为:
for object in all objects:
(1) 获取当前目标类别下所有bbx的信息
(2) 将bbx按照confidence从高到低排序,并记录当前confidence最大的bbx
(3) 计算最大confidence对应的bbx与剩下所有的bbx的IOU,移除所有大于IOU阈值的bbx
(4) 对剩下的bbx,循环执行(2)和(3)直到所有的bbx均满足要求(即不能再移除bbx)
代码实现:
import numpy as np
class Bounding_box:
def __init__(self, x1, y1, x2, y2, score):
self.x1 = x1
self.y1 = y1
self.x2 = x2
self.y2 = y2
self.score = score
def get_iou(boxa, boxb):
max_x = max(boxa.x1, boxb.x1)
max_y = max(boxa.y1, boxb.y1)
min_x = min(boxa.x2, boxb.x2)
min_y = min(boxa.y2, boxb.y2)
if min_x <= max_x or min_y <= max_y:
return 0
area_i = (min_x - max_x) * (min_y - max_y)
area_a = (boxa.x2 - boxa.x1) * (boxa.y2 - boxa.y1)
area_b = (boxb.x2 - boxb.x1) * (boxb.y2 - boxb.y1)
area_u = area_a + area_b - area_i
return float(area_i) / float(area_u)
def NMS(box_lists, k):
box_lists = sorted(box_lists, key=lambda x: x.score, reverse=True)
NMS_lists = [box_lists[0]]
temp_lists = []
for i in range(k):
for j in range(1, len(box_lists)):
iou = get_iou(NMS_lists[i], box_lists[j])
if iou < 0.7:
temp_lists.append(box_lists[j])
if len(temp_lists) == 0:
return NMS_lists
box_lists = temp_lists
temp_lists = []
NMS_lists.append(box_lists[0])
return NMS_lists
box1 = Bounding_box(13, 22, 268, 367, 0.124648176)
box2 = Bounding_box(18, 27, 294, 400, 0.35818103)
box3 = Bounding_box(234, 123, 466, 678, 0.13638769)
box_lists = [box1, box2, box3]
NMS_list = NMS(box_lists, 2)
print NMS_list
print NMS_list[0].x1