tf.contrib.keras中的错误

Question

我在tf_1.2中运行以下代码：

inp = tf.contrib.keras.layers.Input(shape=(height, width, depth)) 
conv_1 = tf.contrib.keras.layers.Convolution2D(conv_depth,(kernel_size, kernel_size), 
    padding='same', 
    kernel_initializer='he_uniform', 
    kernel_regularizer=tf.contrib.keras.regularizers.l2(0.001), 
    activation='relu')(inp)
 model = tf.contrib.keras.models.Model(inputs=inp, outputs=conv_1)
 a = tf.contrib.keras.layers.Input(shape=(height, width, depth))
 enc_a = model(a)

收到以下错误

tensorflow/contrib/keras/python/keras/engine/topology.pyc in run_internal_graph(self, inputs, masks)
1955           # Keep track of unconditional losses
1956           # (e.g. weight regularizers).
-> 1957           self.add_loss(layer.get_losses_for(None), None)
tensorflow/python/layers/base.pyc in add_loss(self, losses, inputs)
254     if not losses:
255       return
--> 256     self._losses += losses
AttributeError: 'Model' object has no attribute '_losses'

如果删除kernel_regularizer，相同的代码也可以工作，如果我导入的是keras而不是tf.contrib.keras