Keras vgg16: Matrixgröße-inkompatibel: In [0]: [16,18432], In [1]: [25088,4096]

Question

Ich benutze Keras mit Tensorflow-Backend, um ein Klassifikationsmodell basierend auf vgg16-Netzwerk zu betreiben. Wenn das Training beginnt, erhalte ich folgende Fehlermeldung: Hier ist der Teil der Kurve ist, dass ich wichtig denken:

Matrix size-incompatible: In[0]: [16,18432], In[1]: [25088,4096] 

und es passiert zwischen diesen beiden Schichten Ich glaube:

flatten_1 (Flatten)    (None, 25088)   0   maxpooling2d_5[0][0] 
____________________________________________________________________________________________________ 
dense_1 (Dense)     (None, 4096)   102764544 flatten_1[0][0] 
__________________________________________________________________________________________________ 

Und hier Spur ist die vollständige Fehler:

Traceback (most recent call last): 
    File "L1.py", line 56, in <module> 
    vgg.fit(batches, valid_batches, nb_epoch=1) 
    File "/vgg16.py", line 220, in fit 
    validation_data=val_batches, nb_val_samples=val_batches.nb_sample) 
    File "..\keras\models.py", line 935, in fit_generator 
    initial_epoch=initial_epoch) 
    File "..\keras\engine\training.py", line 1557, in fit_generator 
    class_weight=class_weight) 
    File "..\keras\engine\training.py", line 1320, in train_on_batch 
    outputs = self.train_function(ins) 
    File "..\keras\backend\tensorflow_backend.py", line 1943, in __ca 
    feed_dict=feed_dict) 
    File "..\tensorflow\python\client\session.py", line 767, in run 
    run_metadata_ptr) 
    File "..\tensorflow\python\client\session.py", line 965, in _run 
    feed_dict_string, options, run_metadata) 
    File "..\tensorflow\python\client\session.py", line 1015, in _do_ 
    target_list, options, run_metadata) 
    File "..\tensorflow\python\client\session.py", line 1035, in _do_ 
    raise type(e)(node_def, op, message) 
tensorflow.python.framework.errors_impl.InvalidArgumentError: Matrix size-incompatible: In[0]: [16,18432], In[1]: [25088,4096] 
     [[Node: MatMul = MatMul[T=DT_FLOAT, transpose_a=false, transpose_b=false, _device="/job:localhost/replica:0/task:0/gpu:0"](Re] 
     [[Node: mul_2/_231 = _Recv[client_terminated=false, recv_device="/job:localhost/replica:0/task:0/cpu:0", send_device="/job:lopu:0", send_device_incarnation=1, tensor_name="edge_667_mul_2", tensor_type=DT_FLOAT, _device="/job:localhost/replica:0/task:0/cpu:0"] 

Caused by op 'MatMul', defined at: 
    File "L1.py", line 51, in <module> 
    vgg = Vgg16() 
    File "/vgg16.py", line 47, in __init__ 
    self.create() 
    File "/vgg16.py", line 139, in create 
    self.FCBlock() 
    File "/vgg16.py", line 113, in FCBlock 
    model.add(Dense(4096, activation='relu')) 
    File "..\keras\models.py", line 332, in add 
    output_tensor = layer(self.outputs[0]) 
    File "..\keras\engine\topology.py", line 572, in __call__ 
    self.add_inbound_node(inbound_layers, node_indices, tensor_indices) 
    File "..\keras\engine\topology.py", line 635, in add_inbound_node 
    Node.create_node(self, inbound_layers, node_indices, tensor_indices) 
    File "..\keras\engine\topology.py", line 166, in create_node 
    output_tensors = to_list(outbound_layer.call(input_tensors[0], mask=input_masks[0])) 
    File "..\keras\layers\core.py", line 814, in call 
    output = K.dot(x, self.W) 
    File "..\keras\backend\tensorflow_backend.py", line 827, in dot 
    out = tf.matmul(x, y) 
    File "..\tensorflow\python\ops\math_ops.py", line 1765, in matmul 
    a, b, transpose_a=transpose_a, transpose_b=transpose_b, name=name) 
    File "..\tensorflow\python\ops\gen_math_ops.py", line 1454, in _m 
    transpose_b=transpose_b, name=name) 
    File "..\tensorflow\python\framework\op_def_library.py", line 763 
    op_def=op_def) 
    File "..\tensorflow\python\framework\ops.py", line 2327, in creat 
    original_op=self._default_original_op, op_def=op_def) 
    File "..\tensorflow\python\framework\ops.py", line 1226, in __ini 
    self._traceback = _extract_stack() 

InvalidArgumentError (see above for traceback): Matrix size-incompatible: In[0]: [16,18432], In[1]: [25088,4096] 
     [[Node: MatMul = MatMul[T=DT_FLOAT, transpose_a=false, transpose_b=false, _device="/job:localhost/replica:0/task:0/gpu:0"](Re] 
     [[Node: mul_2/_231 = _Recv[client_terminated=false, recv_device="/job:localhost/replica:0/task:0/cpu:0", send_device="/job:lopu:0", send_device_incarnation=1, tensor_name="edge_667_mul_2", tensor_type=DT_FLOAT, _device="/job:localhost/replica:0/task:0/cpu:0"] 

Und hier ist die vollständige Modell Zusammenfassung:

____________________________________________________________________________________________________ 
Layer (type)      Output Shape   Param #  Connected to 
==================================================================================================== 
lambda_1 (Lambda)    (None, 3, 226, 226) 0   lambda_input_1[0][0] 
____________________________________________________________________________________________________ 
convolution2d_1 (Convolution2D) (None, 64, 224, 224) 1792  lambda_1[0][0] 
____________________________________________________________________________________________________ 
zeropadding2d_1 (ZeroPadding2D) (None, 64, 226, 226) 0   convolution2d_1[0][0] 
____________________________________________________________________________________________________ 
convolution2d_2 (Convolution2D) (None, 64, 224, 224) 36928  zeropadding2d_1[0][0] 
____________________________________________________________________________________________________ 
maxpooling2d_1 (MaxPooling2D) (None, 64, 112, 112) 0   convolution2d_2[0][0] 
____________________________________________________________________________________________________ 
zeropadding2d_2 (ZeroPadding2D) (None, 64, 114, 114) 0   maxpooling2d_1[0][0] 
____________________________________________________________________________________________________ 
convolution2d_3 (Convolution2D) (None, 128, 112, 112) 73856  zeropadding2d_2[0][0] 
____________________________________________________________________________________________________ 
zeropadding2d_3 (ZeroPadding2D) (None, 128, 114, 114) 0   convolution2d_3[0][0] 
____________________________________________________________________________________________________ 
convolution2d_4 (Convolution2D) (None, 128, 112, 112) 147584  zeropadding2d_3[0][0] 
____________________________________________________________________________________________________ 
maxpooling2d_2 (MaxPooling2D) (None, 128, 56, 56) 0   convolution2d_4[0][0] 
____________________________________________________________________________________________________ 
zeropadding2d_4 (ZeroPadding2D) (None, 128, 58, 58) 0   maxpooling2d_2[0][0] 
____________________________________________________________________________________________________ 
convolution2d_5 (Convolution2D) (None, 256, 56, 56) 295168  zeropadding2d_4[0][0] 
____________________________________________________________________________________________________ 
zeropadding2d_5 (ZeroPadding2D) (None, 256, 58, 58) 0   convolution2d_5[0][0] 
____________________________________________________________________________________________________ 
convolution2d_6 (Convolution2D) (None, 256, 56, 56) 590080  zeropadding2d_5[0][0] 
____________________________________________________________________________________________________ 
zeropadding2d_6 (ZeroPadding2D) (None, 256, 58, 58) 0   convolution2d_6[0][0] 
____________________________________________________________________________________________________ 
convolution2d_7 (Convolution2D) (None, 256, 56, 56) 590080  zeropadding2d_6[0][0] 
____________________________________________________________________________________________________ 
maxpooling2d_3 (MaxPooling2D) (None, 256, 28, 28) 0   convolution2d_7[0][0] 
____________________________________________________________________________________________________ 
zeropadding2d_7 (ZeroPadding2D) (None, 256, 30, 30) 0   maxpooling2d_3[0][0] 
____________________________________________________________________________________________________ 
convolution2d_8 (Convolution2D) (None, 512, 28, 28) 1180160  zeropadding2d_7[0][0] 
____________________________________________________________________________________________________ 
zeropadding2d_8 (ZeroPadding2D) (None, 512, 30, 30) 0   convolution2d_8[0][0] 
____________________________________________________________________________________________________ 
convolution2d_9 (Convolution2D) (None, 512, 28, 28) 2359808  zeropadding2d_8[0][0] 
____________________________________________________________________________________________________ 
zeropadding2d_9 (ZeroPadding2D) (None, 512, 30, 30) 0   convolution2d_9[0][0] 
____________________________________________________________________________________________________ 
convolution2d_10 (Convolution2D) (None, 512, 28, 28) 2359808  zeropadding2d_9[0][0] 
____________________________________________________________________________________________________ 
maxpooling2d_4 (MaxPooling2D) (None, 512, 14, 14) 0   convolution2d_10[0][0] 
____________________________________________________________________________________________________ 
zeropadding2d_10 (ZeroPadding2D) (None, 512, 16, 16) 0   maxpooling2d_4[0][0] 
____________________________________________________________________________________________________ 
convolution2d_11 (Convolution2D) (None, 512, 14, 14) 2359808  zeropadding2d_10[0][0] 
____________________________________________________________________________________________________ 
zeropadding2d_11 (ZeroPadding2D) (None, 512, 16, 16) 0   convolution2d_11[0][0] 
____________________________________________________________________________________________________ 
convolution2d_12 (Convolution2D) (None, 512, 14, 14) 2359808  zeropadding2d_11[0][0] 
____________________________________________________________________________________________________ 
zeropadding2d_12 (ZeroPadding2D) (None, 512, 16, 16) 0   convolution2d_12[0][0] 
____________________________________________________________________________________________________ 
convolution2d_13 (Convolution2D) (None, 512, 14, 14) 2359808  zeropadding2d_12[0][0] 
____________________________________________________________________________________________________ 
maxpooling2d_5 (MaxPooling2D) (None, 512, 7, 7)  0   convolution2d_13[0][0] 
____________________________________________________________________________________________________ 
flatten_1 (Flatten)    (None, 25088)   0   maxpooling2d_5[0][0] 
____________________________________________________________________________________________________ 
dense_1 (Dense)     (None, 4096)   102764544 flatten_1[0][0] 
____________________________________________________________________________________________________ 
dropout_1 (Dropout)    (None, 4096)   0   dense_1[0][0] 
____________________________________________________________________________________________________ 
dense_2 (Dense)     (None, 4096)   16781312 dropout_1[0][0] 
____________________________________________________________________________________________________ 
dropout_2 (Dropout)    (None, 4096)   0   dense_2[0][0] 
____________________________________________________________________________________________________ 
dense_3 (Dense)     (None, 1000)   4097000  dropout_2[0][0] 
==================================================================================================== 
Total params: 138,357,544 
Trainable params: 138,357,544 
Non-trainable params: 0 
_________________________________________________ 

Ich verstehe, dass es Matrizen mit Größen inkompatibel für Matrixpunkt Multiplikation gibt, aber wie kann ich das beheben?

Answer 1

Die Reihenfolge der Bildkanäle scheint eher für Theano als für Tensorflow zu gelten. Korrekte Reihenfolge in model.summary sollte wie folgt aussehen:

input_1 (InputLayer)   (None, 224, 224, 3)  0   
_________________________________________________________________ 
block1_conv1 (Conv2D)  (None, 224, 224, 64)  1792  
_________________________________________________________________ 
block1_conv2 (Conv2D)  (None, 224, 224, 64)  36928  

… 

block5_pool (MaxPooling2D) (None, 7, 7, 512)   0   
_________________________________________________________________ 
flatten (Flatten)   (None, 25088)    0   
_________________________________________________________________ 
fc1 (Dense)     (None, 4096)    102764544 

Im letzten MaxPool Sie haben 512, 7, 7 aber es sollte 7, 7, 512 sein.

Stellen Sie sicher, Tensorflow verwenden eher dann Theano Bestellung durch diese Zeilen Code hinzufügen, bevor Sie Ihr Modell kompilieren:

from keras import backend as K  
K.set_image_dim_ordering('tf') 

Sie können auch die Keras Konfigurationsdatei hier zu: ~/.keras/keras.json

Was sollte enthalten:

{ 
"epsilon": 1e-07, 
"floatx": "float32", 
"backend": "tensorflow", 
"image_dim_ordering": "tf" 
}