Trjtdtk

This is probably a very stupid question but the regular confusion of arrays and tensors in speech makes me unable to find any answer on the net.

If have a samples long array of "tensorflow.python.framework.ops.Tensor" which are indeed correct numpy arrays of size (299,299,3) once evaluated. My model (inceptionv3) accepts default input of size (299,299,3). Why does it only recognize my input as a (samples, 1) array ? Do I really need to evaluate each tensor ? As evaluating one takes like 5s ...

How can I change the input of the model to get the model working ?
Or how can I change my tensors in a reasonable time to the right format?

I have :

train_images.shape Out[170]: (24,)

Precisely:

train_images
Out[167]:
[<tf.Tensor 'Squeeze_3:0' shape=(299, 299, 3) dtype=float32>,
<tf.Tensor 'Squeeze_4:0' shape=(299, 299, 3) dtype=float32>,
<tf.Tensor 'Squeeze_5:0' shape=(299, 299, 3) dtype=float32>,
<tf.Tensor 'Squeeze_6:0' shape=(299, 299, 3) dtype=float32>,

And:

type(train_images[1]) Out[168]: tensorflow.python.framework.ops.Tensor

Obviously gets :

ValueError: Error when checking input: expected input_5 to have 4 dimensions, but got array with shape (24, 1)

Here is my model code (it's basic out of the box)

#%%
train_labels=augmented_label
train_images=augmented_ims
# Importing the keras libraries
import keras
from keras.models import Model
from keras.layers import Dense, GlobalAveragePooling2D, Flatten, BatchNormalization, Activation, Dropout
from keras.callbacks import ModelCheckpoint, TensorBoard
from keras.applications.inception_v3 import InceptionV3
# Pre-build model

base_model = InceptionV3(include_top = False, weights = None,)
 # Adding output layers
x = base_model.output
x = GlobalAveragePooling2D()(x)
output = Dense(units = 2, activation = 'sigmoid')(x)
 # Creating the whole model
inception_model = Model(base_model.input, output)

 # Summary of the model
 #inception_model.summary()

 # Compiling the model
inception_model.compile(optimizer = keras.optimizers.Adam(lr = 0.001), 
 loss = 'categorical_crossentropy', 
 metrics = ['accuracy'])
#%%
from keras.utils.np_utils import to_categorical
train_labels=to_categorical(np.asarray(augmented_label))
train_images=augmented_ims
train_images=np.asarray(train_images)
train_images=train_images/255
inception_model.fit(train_images, train_labels, epochs=5)

Thx