Keras自定义IOU方式

时间:2022-09-13 07:30:48

我就废话不多说了,大家还是直接看代码吧!

?
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
def iou(y_true, y_pred, label: int):
  """
  Return the Intersection over Union (IoU) for a given label.
  Args:
    y_true: the expected y values as a one-hot
    y_pred: the predicted y values as a one-hot or softmax output
    label: the label to return the IoU for
  Returns:
    the IoU for the given label
  """
  # extract the label values using the argmax operator then
  # calculate equality of the predictions and truths to the label
  y_true = K.cast(K.equal(K.argmax(y_true), label), K.floatx())
  y_pred = K.cast(K.equal(K.argmax(y_pred), label), K.floatx())
  # calculate the |intersection| (AND) of the labels
  intersection = K.sum(y_true * y_pred)
  # calculate the |union| (OR) of the labels
  union = K.sum(y_true) + K.sum(y_pred) - intersection
  # avoid divide by zero - if the union is zero, return 1
  # otherwise, return the intersection over union
  return K.switch(K.equal(union, 0), 1.0, intersection / union)
 
def mean_iou(y_true, y_pred):
  """
  Return the Intersection over Union (IoU) score.
  Args:
    y_true: the expected y values as a one-hot
    y_pred: the predicted y values as a one-hot or softmax output
  Returns:
    the scalar IoU value (mean over all labels)
  """
  # get number of labels to calculate IoU for
  num_labels = K.int_shape(y_pred)[-1] - 1
  # initialize a variable to store total IoU in
  mean_iou = K.variable(0)
  
  # iterate over labels to calculate IoU for
  for label in range(num_labels):
    mean_iou = mean_iou + iou(y_true, y_pred, label)
    
  # divide total IoU by number of labels to get mean IoU
  return mean_iou / num_labels

补充知识:keras 自定义评估函数和损失函数loss训练模型后加载模型出现ValueError: Unknown metric function:fbeta_score

keras自定义评估函数

有时候训练模型,现有的评估函数并不足以科学的评估模型的好坏,这时候就需要自定义一些评估函数,比如样本分布不均衡是准确率accuracy评估无法判定一个模型的好坏,这时候需要引入精确度和召回率作为评估标准,不幸的是keras没有这些评估函数。

以下是参考别的文章摘取的两个自定义评估函数

召回率:

?
1
2
3
4
5
def recall(y_true, y_pred):
  true_positives = K.sum(K.round(K.clip(y_true * y_pred, 0, 1)))
  possible_positives = K.sum(K.round(K.clip(y_true, 0, 1)))
  recall = true_positives / (possible_positives + K.epsilon())
  return recall

精确度:

?
1
2
3
4
5
def precision(y_true, y_pred):
  true_positives = K.sum(K.round(K.clip(y_true * y_pred, 0, 1)))
  predicted_positives = K.sum(K.round(K.clip(y_pred, 0, 1)))
  precision = true_positives / (predicted_positives + K.epsilon())
  return precision

自定义了评估函数,一般在编译模型阶段加入即可:

model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy', precision, recall])

自定义了损失函数focal_loss一般也在编译阶段加入:

model.compile(optimizer=Adam(lr=0.0001), loss=[focal_loss],
metrics=['accuracy',fbeta_score], )

其他的没有特别要注意的点,直接按照原来的思路训练一版模型出来就好了,关键的地方在于加载模型这里,自定义的函数需要特殊的加载方式,不然会出现加载没有自定义函数的问题:ValueError: Unknown loss function:focal_loss

解决方案:

?
1
2
3
model_name = 'test_calssification_model.h5'
model_dfcw = load_model(model_name,
            custom_objects={'focal_loss': focal_loss,'fbeta_score':fbeta_score})

注意点:将自定义的损失函数和评估函数都加入到custom_objects里,以上就是在自定义一个损失函数从编译模型阶段到加载模型阶段出现的所有的问题。

以上这篇Keras自定义IOU方式就是小编分享给大家的全部内容了,希望能给大家一个参考,也希望大家多多支持服务器之家。

原文链接:https://blog.csdn.net/xijuezhu8128/article/details/86608333