原始参考:https://blog.csdn.net/jason__sz/article/details/79179107
感觉原文注解有些问题,因此重新整理注解了一下,具体见下面:
import numpy as np
def py_cpu_nms(dets, thresh):
"""Pure Python NMS baseline."""
# 所有box的坐标信息
x1 = dets[:, 0]
y1 = dets[:, 1]
x2 = dets[:, 2]
y2 = dets[:, 3]
scores = dets[:, 4]
areas = (x2 - x1 + 1) * (y2 - y1 + 1) # 计算出所有box的面积
order = scores.argsort()[::-1] # 图片评分按降序排序
keep = [] # 用来存放保留的box的索引
while order.size > 0:
i = order[0] # i 是还未处理的图片中的最大评分索引
keep.append(i) # 保留改图片的值
# 矩阵操作,下面计算的是box_i分别与其余box相交的矩形的坐标
xx1 = np.maximum(x1[i], x1[order[1:]])
yy1 = np.maximum(y1[i], y1[order[1:]])
xx2 = np.minimum(x2[i], x2[order[1:]])
yy2 = np.minimum(y2[i], y2[order[1:]])
# 计算出各个相交矩形的面积
w = np.maximum(0.0, xx2 - xx1 + 1)
h = np.maximum(0.0, yy2 - yy1 + 1)
inter = w * h
# 计算重叠比例
ovr = inter / (areas[i] + areas[order[1:]] - inter)
#只保留比例小于阙值的box,然后继续处理,因为这可能是另外一个目标
inds = np.where(ovr <= thresh)[0]
order = order[inds + 1]
return keep