código
"""
@author: hukai
@Time: 2022/4/6 15:04
confusionMetric # 注意：此处横着代表预测值，竖着代表真实值，与之前介绍的相反
P\L P N
P TP FP
N FN TN
"""
import numpy as np
class SegmentationMetric(object):
def __init__(self, num_classes):
self.num_classes = num_classes
self.confusionMatrix = np.zeros((self.num_classes, self.num_classes))
def updateConfusionMatrix(self, predict, label):
mask = (label >= 0) & (label < self.num_classes) # 寻找target中为目标的像素索引
new_label = self.num_classes * label[mask] + predict[mask] # 二维的predict拉到一维
count = np.bincount(new_label, minlength=self.num_classes**2)
self.confusionMatrix += count.reshape(self.num_classes, self.num_classes)
return self.confusionMatrix
def pixelAccuracy(self):
acc = np.diag(self.confusionMatrix).sum() / self.confusionMatrix.sum()
return acc
def classPixelAccuracy(self):
# 横着代表预测值，竖着代表真实值
classAcc = np.diag(self.confusionMatrix) / self.confusionMatrix.sum(axis=1)
# 返回的是一个列表值，如：[0.90, 0.80, 0.96]，表示类别1 2 3各类别的预测准确率
return classAcc
def meanPixelAccuracy(self):
classAcc = self.classPixelAccuracy()
meanAcc = np.nanmean(classAcc) # 求平均值，遇到Nan类型，其值取为0
return meanAcc
def meanIOU(self):
intersection = np.diag(self.confusionMatrix) # 对角线元素
union = np.sum(self.confusionMatrix, axis=1) + np.sum(self.confusionMatrix, axis=0) - np.diag(self.confusionMatrix)
IOU = intersection / union
mIOU = np.nanmean(IOU)
return mIOU
if __name__ == '__main__':
Predict = np.array([[0, 0, 1, 0, 2, 2], [0, 1, 1, 1, 2, 2]]) # 可直接换成预测图片
Label = np.array([[0, 0, 1, 1, 2, 2], [0, 0, 1, 1, 2, 2]]) # 可直接换成标注图片
metric = SegmentationMetric(3) # 3表示有3个分类，有几个分类就填几
confusionMatrix = metric.updateConfusionMatrix(Predict, Label)
pa = metric.pixelAccuracy()
cpa = metric.classPixelAccuracy()
mpa = metric.meanPixelAccuracy()
mIoU = metric.meanIOU()
print(confusionMatrix)
print('pa is : %f' % pa)
print('cpa is :') # 列表
print(cpa)
print('mpa is : %f' % mpa)
print('mIoU is : %f' % mIoU)
resultado:
