Hi Saket,

you have to be careful with input/output shapes of your data. Take this example as some solution that may help you get a sense how to manipulate with forms of your data:

import numpy as np

train_x = np.expand_dims(train_x, axis=2)

inp = Input(shape=(len(train_x[0]), 1))

conv1 = Conv1D(filters=80, kernel_size=15)(inp)

conv2 = Conv1D(filters=60, kernel_size=10)(conv1)

pool2 = MaxPool1D(pool_size=60)(conv2)

flat = Flatten()(pool2)

drop = Dropout(0.2)(flat)

dense1 = Dense(100, activation='tanh')(drop)

reshape = Reshape([len(train_x[0]),1])(dense1)

model = Model(inp, reshape)

model.compile(loss='mse', optimizer='rmsprop')

...

by np.expand_dims you can increase the dimension of your dataset and also making use of Reshape layer (in Keras) you can manipulate its shape in the neural network.

https://www.tensorflow.org/api_docs/python/tf/keras/layers/Reshape

regards,

Hi Saket,

Tomas is right you need to consider shape of your input. Also see if input layer in pretrained model has dimensions consistent with your data.

Let's analyze your error step by step.

ValueError: Error when checking input: expected image_input to have 4 dimensions, but got array with shape (224, 224, 3)

Your pretrained model must have been trained batch wise. Your model is expecting input of shape (batch_size, 224, 224, 3) or it could also be a transpose of same. See whether pretrained keras model is with shape mode 'th' or not.

For a single image you can just reshape into a tensor of size (1, 224, 224, 3) As suggested by Tomas.

ValueError: The shape of the input to "Flatten" is not fully defined (got (None, None, 512). Make sure to pass a complete "input_shape"

This error is because you are not using pretrained model correctly. To build a model on top of pretrained model. First build a pretrained network and load the weights. Now define a separate classifier to be used on top of pretrained network(separate model).Add this classifier to your pretrained model using .add. Here dimension should be consistent.

ValueError : Error when checking target (output shape error)

When you stack two models output shape of one should be input of other.

A small example (not tested just to give an overview):

from keras import applications

model = applications.VGG16(weights='imagenet', include_top=False)

new_classifier=Sequential()

new_classifier.add(Flatten(input_shape=model.output_shape[1:]))

new_classifier.add(Dense(1))

model.add(new_classifier)

model.compile(loss='binary_crossentropy', metrics=[dice_coef],

optimizer='adam', class_mode="sigmoid")

Now for using this model for training. Input dimension should be consistent with the input layer of pretrained model. If VGG16 takes image of size (224, 224, 3) it will not work for image of size (256,256,3).

from keras.preprocessing import image

from keras.applications.vgg16 import preprocess_input

img = image.load_img("image.jpg", target_size=(224, 224))

img = image.img_to_array(img) img = np.expand_dims(img, axis=0)

img = preprocess_input(img)

Now you can fit your model on img and likewise can do predictions.

Please have a look at following link for using pretrained model with keras.

https://keras.io/applications/

https://gist.github.com/fchollet/7eb39b44eb9e16e59632d25fb3119975

Hello Tomas Vantuch & Mohammad Asif Khan ,

Thank you for the suggestions. I tried doing that way with the following code:

train_x = np.expand_dims(train_X, axis=2)

train_y = np.expand_dims(train_Y, axis=2)

print(train_X.shape) # output - (670, 224, 224, 3)

print(train_Y.shape) # output - (670, 224, 224, 1)

print(train_x.shape) # output - (670, 224, 1, 224, 3)

print(train_y.shape) # output - (670, 224, 1, 224, 1)

def vgg16_(IMG_WIDTH=224,IMG_HEIGHT=224,IMG_CHANNELS=3):

inputs = Input(shape=(len(train_x[0]), 1))

x = Conv2D(64, (3, 3), activation='relu', padding='same', name='block1_conv1')(inputs)

x = Conv2D(64, (3, 3), activation='relu', padding='same', name='block1_conv2')(x)

x = MaxPooling2D((2, 2), strides=(2, 2), name='block1_pool')(x)

# Block 2

x = Conv2D(128, (3, 3), activation='relu', padding='same', name='block2_conv1')(x)

x = Conv2D(128, (3, 3), activation='relu', padding='same', name='block2_conv2')(x)

x = MaxPooling2D((2, 2), strides=(2, 2), name='block2_pool')(x)

# Block 3

x = Conv2D(256, (3, 3), activation='relu', padding='same', name='block3_conv1')(x)

x = Conv2D(256, (3, 3), activation='relu', padding='same', name='block3_conv2')(x)

x = Conv2D(256, (3, 3), activation='relu', padding='same', name='block3_conv3')(x)

x = MaxPooling2D((2, 2), strides=(2, 2), name='block3_pool')(x)

# Block 4

x = Conv2D(512, (3, 3), activation='relu', padding='same', name='block4_conv1')(x)

x = Conv2D(512, (3, 3), activation='relu', padding='same', name='block4_conv2')(x)

x = Conv2D(512, (3, 3), activation='relu', padding='same', name='block4_conv3')(x)

x = MaxPooling2D((2, 2), strides=(2, 2), name='block4_pool')(x)

# Block 5

x = Conv2D(512, (3, 3), activation='relu', padding='same', name='block5_conv1')(x)

x = Conv2D(512, (3, 3), activation='relu', padding='same', name='block5_conv2')(x)

x = Conv2D(512, (3, 3), activation='relu', padding='same', name='block5_conv3')(x)

x = MaxPooling2D((2, 2), strides=(2, 2), name='block5_pool')(x)

x = Flatten()(x)

x = Dropout(0.2)(x)

x = Dense(100, activation='tanh')(x)

x = Reshape([len(train_x[0]),1])(x)

model = Model(inputs, reshape)

model.compile(loss='mse', optimizer='rmsprop')

return model

But, Unfortunately I am getting this error by fitting this model on training data

ValueError: Input 0 is incompatible with layer block1_conv1: expected ndim=4, found ndim=3

What should I do now? Thank you in advance.

you are trying to pass image with shape (224,224,3) where your network is designed to take input image with shape (224,224,3,n) where n refers to number of pictures,you need to tell specifically to your network "how many pictures you are passing",this is why the input shape of your designed network is 4 but as you are only passing (224,224,3) the network refuses to work because you are passing 3d input and not telling how many pictures you are passing,i have worked previously on resnet50 and vgg16,i can share you my code but you need to improve those codes(for your particular problem) :

1. https://github.com/mobassir94/Practicing-Tensorflow/blob/master/kaggle_Pneumonia_Detection.ipynb

2. https://github.com/mobassir94/Practicing-Tensorflow/blob/master/kaggle_pneumonia(overfitted).ipynb

Question : How to implement Pre-trained models with only modifications in the output layer?

Answer : download weight of pre-trained model like resnet50 or vgg16 then delete the last layer of those models and freeze all layers by saying model.trainable = False(if you want to make some layers trainable then you can make that particular layer trainable by saying : model.trainable = True) then add dense layer at the end following output layer(sigmoid or softmax depending on your problem)

Thanks Syed Mobassir Hossen.

Check out this link:

https://github.com/GeorgeSeif/Semantic-Segmentation-Suite