TensorFlow实现center loss

TensorFlow实现center loss：Center loss是ECCV2016中一篇论文提出来的概念，主要思想就是在softmax loss基础上额外加入一个正则项，让网络中每一类样本的特征向量都能够尽量聚在一起。

具体的原理推导等请参考论文，论文作者放出了Caffe实现，网上还能找到mxnet的实现，这里我放出一个TensorFlow版的实现及详细注释，代码很短，如下：

def get_center_loss(features, labels, alpha, num_classes):
    # alpha:中心的更新比例
    # 获取特征长度
    len_features = features.get_shape()[ 1 ]
    # 建立一个变量，存储每一类的中心，不训练 9
    centers = tf.get_variable( 'centers' , [num_classes, len_features], dtype=tf.float32,
        initializer=tf.constant_initializer( 0 ), trainable=False)
    # 将特征reshape成一维
    labels = tf.reshape(labels, [- 1 ])

    # 获取当前batch每个样本对应的中心
    centers_batch = tf.gather(centers, labels)
    # 计算center loss的数值
    loss = tf.nn.l2_loss(features - centers_batch)

    # 以下为更新中心的步骤
    diff = centers_batch - features

    # 获取一个batch中同一样本出现的次数，这里需要理解论文中的更新公式
    unique_label, unique_idx, unique_count = tf.unique_with_counts(labels)
    appear_times = tf.gather(unique_count, unique_idx)
    appear_times = tf.reshape(appear_times, [- 1 , 1 ])

    diff = diff / tf.cast(( 1 + appear_times), tf.float32)
    diff = alpha * diff
    # 更新中心
    centers = tf.scatter_sub(centers, labels, diff)

    return loss, centers


center loss代码注释(caffe新添加层)
http://blog.csdn.net/liyuan123zhouhui/article/details/60139981