Question

我有一个Keras自动编码器模型，它采用MNIST图像和形状矢量（(12,)矢量）。我试图将形状向量添加到解码阶段获得的要素图中（参见https://arxiv.org/pdf/1603.02199.pdf图4中的空间平铺）。该模型产生了与lambda层相关的奇怪错误。我无法从追溯中理解错误的原因。

源代码：

input_img = Input(shape=(28, 28, 1))
shape_input = Input(shape=(12,))
shape_input_reshaped = Reshape((1, 1, 12))(shape_input)

x = Conv2D(16, (3, 3), activation='relu', padding='same')(input_img)
x = MaxPooling2D((2, 2), padding='same')(x)
x = Conv2D(12, (3, 3), activation='relu', padding='same')(x)
x = MaxPooling2D((2, 2), padding='same')(x)
x = Conv2D(12, (3, 3), activation='relu', padding='same')(x)
encoded = MaxPooling2D((2, 2), padding='same')(x)

# at this point the representation is (4, 4, 8) i.e. 128-dimensional
x = Conv2D(12, (3, 3), activation='relu', padding='same')(encoded)
shape_vec = Lambda(lambda x: K.tile(x, [-1, 4, 4, 1]))(shape_input_reshaped)
x = Add()([x, shape_vec])
x = UpSampling2D((2, 2))(x)
x = Conv2D(12, (3, 3), activation='relu', padding='same')(x)
shape_vec = Lambda(lambda x: K.tile(x, [-1, 8, 8, 1]))(shape_input_reshaped)
x = Add()([x, shape_vec])
x = UpSampling2D((2, 2))(x)
x = Conv2D(16, (3, 3), activation='relu')(x)
x = UpSampling2D((2, 2))(x)
decoded = Conv2D(1, (3, 3), activation='sigmoid', padding='same')(x)

autoencoder = Model(inputs=[input_img, shape_input], outputs=decoded)
autoencoder.compile(optimizer='adadelta', loss='binary_crossentropy')

错误日志：

InvalidArgumentError                      Traceback (most recent call last)
<ipython-input-8-adda46da1298> in <module>()
      5                 shuffle=True,
      6                 validation_data=([x_test, test_shapes], x_test_transformed),
----> 7                 callbacks=[TensorBoard(log_dir='./logs/2')])
      8 

/media/zero/data1/anaconda3/lib/python3.6/site-packages/keras/engine/training.py in fit(self, x, y, batch_size, epochs, verbose, callbacks, validation_split, validation_data, shuffle, class_weight, sample_weight, initial_epoch, steps_per_epoch, validation_steps, **kwargs)
   1703                               initial_epoch=initial_epoch,
   1704                               steps_per_epoch=steps_per_epoch,
-> 1705                               validation_steps=validation_steps)
   1706 
   1707     def evaluate(self, x=None, y=None,

/media/zero/data1/anaconda3/lib/python3.6/site-packages/keras/engine/training.py in _fit_loop(self, f, ins, out_labels, batch_size, epochs, verbose, callbacks, val_f, val_ins, shuffle, callback_metrics, initial_epoch, steps_per_epoch, validation_steps)
   1233                         ins_batch[i] = ins_batch[i].toarray()
   1234 
-> 1235                     outs = f(ins_batch)
   1236                     if not isinstance(outs, list):
   1237                         outs = [outs]

/media/zero/data1/anaconda3/lib/python3.6/site-packages/keras/backend/tensorflow_backend.py in __call__(self, inputs)
   2476         session = get_session()
   2477         updated = session.run(fetches=fetches, feed_dict=feed_dict,
-> 2478                               **self.session_kwargs)
   2479         return updated[:len(self.outputs)]
   2480 

/media/zero/data1/anaconda3/lib/python3.6/site-packages/tensorflow/python/client/session.py in run(self, fetches, feed_dict, options, run_metadata)
    903     try:
    904       result = self._run(None, fetches, feed_dict, options_ptr,
--> 905                          run_metadata_ptr)
    906       if run_metadata:
    907         proto_data = tf_session.TF_GetBuffer(run_metadata_ptr)

/media/zero/data1/anaconda3/lib/python3.6/site-packages/tensorflow/python/client/session.py in _run(self, handle, fetches, feed_dict, options, run_metadata)
   1138     if final_fetches or final_targets or (handle and feed_dict_tensor):
   1139       results = self._do_run(handle, final_targets, final_fetches,
-> 1140                              feed_dict_tensor, options, run_metadata)
   1141     else:
   1142       results = []

/media/zero/data1/anaconda3/lib/python3.6/site-packages/tensorflow/python/client/session.py in _do_run(self, handle, target_list, fetch_list, feed_dict, options, run_metadata)
   1319     if handle is None:
   1320       return self._do_call(_run_fn, feeds, fetches, targets, options,
-> 1321                            run_metadata)
   1322     else:
   1323       return self._do_call(_prun_fn, handle, feeds, fetches)

/media/zero/data1/anaconda3/lib/python3.6/site-packages/tensorflow/python/client/session.py in _do_call(self, fn, *args)
   1338         except KeyError:
   1339           pass
-> 1340       raise type(e)(node_def, op, message)
   1341 
   1342   def _extend_graph(self):

InvalidArgumentError: Expected multiples[0] >= 0, but got -1
     [[Node: lambda_1/Tile = Tile[T=DT_FLOAT, Tmultiples=DT_INT32, _device="/job:localhost/replica:0/task:0/device:GPU:0"](reshape_1/Reshape, lambda_1/Tile/multiples)]]
     [[Node: loss/mul/_213 = _Recv[client_terminated=false, recv_device="/job:localhost/replica:0/task:0/device:CPU:0", send_device="/job:localhost/replica:0/task:0/device:GPU:0", send_device_incarnation=1, tensor_name="edge_1303_loss/mul", tensor_type=DT_FLOAT, _device="/job:localhost/replica:0/task:0/device:CPU:0"]()]]

Answer 1

我通过在K.tile中将-1替换为1来训练你的模型。代码片段：

x = Conv2D(12, (3, 3), activation='relu', padding='same')(encoded)
shape_vec = Lambda(lambda x: K.tile(x, [1, 4, 4, 1]))(shape_input_reshaped)
x = Add()([x, shape_vec])
x = UpSampling2D((2, 2))(x)
x = Conv2D(12, (3, 3), activation='relu', padding='same')(x)
shape_vec = Lambda(lambda x: K.tile(x, [1, 8, 8, 1]))(shape_input_reshaped)
x = Add()([x, shape_vec])
x = UpSampling2D((2, 2))(x)
x = Conv2D(16, (3, 3), activation='relu')(x)
x = UpSampling2D((2, 2))(x)
decoded = Conv2D(1, (3, 3), activation='sigmoid', padding='same')(x)

Keras lambda：InvalidArgumentError：预期的倍数[0]＆gt; = 0，但得到-1

1 个答案: