机器学习笔记 - 关于Contrastive Loss对比损失

一、对比损失

虽然二元交叉熵（下图公式）肯定是损失函数的有效选择，但它不是唯一的选择（甚至不是最佳选择）。

然而，实际上有一个更适合孪生网络的损失函数，称为对比损失。

其中Y是我们的标签。如果图像对属于同一类，则为 1，如果图像对属于不同类，则为 0。

$D_{w}$ 变量是孪生网络的输出之间的欧几里得距离。

max 函数取最大值 0 和边距 m 减去距离。

孪生网络的目标不是对一组图像对进行分类，而是区分它们。从本质上讲，对比损失是评估孪生网络在图像对之间的区分程度。

如果能读得进去论文，可以更深入了解下。http://yann.lecun.com/exdb/publis/pdf/hadsell-chopra-lecun-06.pdfhttp://yann.lecun.com/exdb/publis/pdf/hadsell-chopra-lecun-06.pdf 在 Keras/TensorFlow 代码中实现的对比损失函数的公式（也是论文中说的exact loss functi）：

二、参考代码

Image similarity estimation using a Siamese Network with a contrastive lossKeras documentationhttps://keras.io/examples/vision/siamese_contrastive/ 摘抄函数定义1

def loss(margin=1):"""Provides 'constrastive_loss' an enclosing scope with variable 'margin'.Arguments:margin: Integer, defines the baseline for distance for which pairsshould be classified as dissimilar. - (default is 1).Returns:'constrastive_loss' function with data ('margin') attached."""# Contrastive loss = mean( (1-true_value) * square(prediction) +#                         true_value * square( max(margin-prediction, 0) ))def contrastive_loss(y_true, y_pred):"""Calculates the constrastive loss.Arguments:y_true: List of labels, each label is of type float32.y_pred: List of predictions of same length as of y_true,each label is of type float32.Returns:A tensor containing constrastive loss as floating point value."""square_pred = tf.math.square(y_pred)margin_square = tf.math.square(tf.math.maximum(margin - (y_pred), 0))return tf.math.reduce_mean((1 - y_true) * square_pred + (y_true) * margin_square)return contrastive_loss

摘抄函数定义2

# import the necessary packages
import tensorflow.keras.backend as K
import tensorflow as tf
def contrastive_loss(y, preds, margin=1):# explicitly cast the true class label data type to the predicted# class label data type (otherwise we run the risk of having two# separate data types, causing TensorFlow to error out)y = tf.cast(y, preds.dtype)# calculate the contrastive loss between the true labels and# the predicted labelssquaredPreds = K.square(preds)squaredMargin = K.square(K.maximum(margin - preds, 0))loss = K.mean(y * squaredPreds + (1 - y) * squaredMargin)# return the computed contrastive loss to the calling functionreturn loss