tensorflow对模型的权重进行l2正则化(l2 regularization/weight decay)

这里仅对卷积的权重进行l2正则化,biases的权重可以通过根据名称搜索biases的变量，添加变量到regularized_weights即可

regularized_weights=[]
for val in tf.global_variables():
    if 'conv_3d/w:0' in val.name:
        regularized_weights.append(val)
regularizer = tf.contrib.layers.l2_regularizer(0.00004)
regularization_loss = tf.contrib.layers.apply_regularization(regularizer=regularizer,weights_list=regularized_weights)