本文共 3260 字，大约阅读时间需要 10 分钟。
我们来解读一下，中心损失，再来看代码。
链接：
我们的重点是分析代码，所以定义部分，大家详情参见上面的博客。
代码：
#coding=gbk'''Created on 2020年4月20日@author: DELL'''import tensorflow as tfimport numpy as npdata = [[1,1,1,1,1], [1,1,2,1,1], [1,1,3,1,1], [1,1,4,1,1], [2,2,2,1,2], [2,2,2,2,2], [2,2,2,3,2], [3,3,3,3,1], [3,3,3,3,2]]label = [0,0,0,0,1,1,1,2,2]data = np.array(data,dtype = 'float32')label = np.array(label)data = tf.convert_to_tensor(data)label = tf.convert_to_tensor(label)def center_loss(features, label, alfa, nrof_classes): """Center loss based on the paper "A Discriminative Feature Learning Approach for Deep Face Recognition" (http://ydwen.github.io/papers/WenECCV16.pdf) """ nrof_features = features.get_shape()[1] centers = tf.get_variable('centers', [nrof_classes, nrof_features], dtype=tf.float32,initializer=tf.constant_initializer(0), trainable=False) #定义一个全零的centers, [nrof_classes, nrof_features]->(类别数，特征维度) #print(sess.run(centers)) label = tf.reshape(label, [-1]) #一维向量 centers_batch = tf.gather(centers, label) #[batch_size,nrof_features] #按照label将centers归类，形成的新矩阵维度为 [label_size,nrof_features] diff = (1 - alfa) * (centers_batch - features) #乘上我们的因子alfa [label_size,nrof_features] centers = tf.scatter_sub(centers, label, diff) #按照label用centers - diff，产生本次的centers with tf.control_dependencies([centers]):#注意这个函数的作用，是限制计算顺序的，即先计算centers，在利用计算好的centers去计算centers_batch以求loss loss = tf.reduce_mean(tf.square(features - centers_batch)) return loss, centers,features,centers_batch,features - centers_batchloss, cen, fea, cen_bat,a = center_loss(data,label,0.5,3)sess = tf.Session()init = tf.global_variables_initializer()sess.run(init)print(sess.run(cen))#print(sess.run(loss))print(sess.run(fea))#print(sess.run(cen_bat))print(sess.run(a))print(sess.run(fea - cen_bat))print(sess.run(tf.square(fea - cen_bat)))print(sess.run(loss)) '''验证tf.scatter_sub函数sess = tf.Session()ref = tf.Variable([1, 2, 3],dtype = tf.int32)indices = tf.constant([0, 0, 1, 1],dtype = tf.int32)updates = tf.constant([9, 10, 11, 12],dtype = tf.int32)sub = tf.scatter_sub(ref, indices, updates)with tf.Session() as sess: sess.run(tf.global_variables_initializer()) print (sess.run(sub))''' 结果：
1.centers:[[2. 2. 5. 2. 2. ] [3. 3. 3. 3. 3. ] [3. 3. 3. 3. 1.5]]2.features:[[1. 1. 1. 1. 1.] [1. 1. 2. 1. 1.] [1. 1. 3. 1. 1.] [1. 1. 4. 1. 1.] [2. 2. 2. 1. 2.] [2. 2. 2. 2. 2.] [2. 2. 2. 3. 2.] [3. 3. 3. 3. 1.] [3. 3. 3. 3. 2.]]3.centers_batch[[2. 2. 5. 2. 2. ] [2. 2. 5. 2. 2. ] [2. 2. 5. 2. 2. ] [2. 2. 5. 2. 2. ] [3. 3. 3. 3. 3. ] [3. 3. 3. 3. 3. ] [3. 3. 3. 3. 3. ] [3. 3. 3. 3. 1.5] [3. 3. 3. 3. 1.5]]4.features - centers_batch[[-1. -1. -4. -1. -1. ] [-1. -1. -3. -1. -1. ] [-1. -1. -2. -1. -1. ] [-1. -1. -1. -1. -1. ] [-1. -1. -1. -2. -1. ] [-1. -1. -1. -1. -1. ] [-1. -1. -1. 0. -1. ] [ 0. 0. 0. 0. -0.5] [ 0. 0. 0. 0. 0.5]]5.loss1.4111111
主要用到的函数：1.tf.gather(data,labels),将data按labels扩充
2.tf.scatter_sub(data,label,data_1),按label用data - data_
3.with tf.control_dependencies(): ,限制运算顺序
在实验验证时注意的点是：不要多次sess.run()某个张量涉及到带有依赖关系的张量，比如这里的loss，计算loss时 会 主动更新一次值，导致运算结果出错。原理我还没搞清，日后补上
转载地址：http://lmten.baihongyu.com/