Myujth

I am doing a keras application that predicts a group of coordinates from a 3D Image Input. My architecture is the following :

img_input = keras.layers.Input((96,64,64,1))

x = keras.layers.Conv3D(32, (3, 3 ,3), activation='relu', padding='same', name='block1_conv1',trainable=True)(img_input)
x = keras.layers.Conv3D(32, (3, 3 ,3), activation='relu', padding='same', name='block1_conv2',trainable=True)(x)
x = keras.layers.MaxPooling3D((2, 2, 2), name='block1_pool')(x)

# Block 2
x = keras.layers. Conv3D(64, (3, 3 ,3), activation='relu', padding='same', name='block2_conv1',trainable=True)(x)
x = keras.layers.Conv3D(64, (3, 3 ,3), activation='relu', padding='same', name='block2_conv2',trainable=True)(x)
x = keras.layers.MaxPooling3D((2, 2 ,2), name='block2_pool')(x)

# Block 3
x = keras.layers.Conv3D(128, (3, 3 ,3), activation='relu', padding='same', name='block3_conv1',trainable=True)(x)
x = keras.layers.Conv3D(128, (3, 3 , 3), activation='relu', padding='same', name='block3_conv2',trainable=True)(x)
x = keras.layers.MaxPooling3D((2, 2 ,2), name='block3_pool')(x)

# Block 3
x = keras.layers.Conv3D(128, (3, 3 ,3), activation='relu', padding='same', name='block4_conv1',trainable=True)(x)
x = keras.layers.Conv3D(128, (3, 3 , 3), activation='relu', padding='same', name='block4_conv2',trainable=True)(x)
x = keras.layers.MaxPooling3D((2, 2 ,2), name='block4_pool')(x)

x = keras.layers.Flatten(name='flatten')(x)

x = keras.layers.Dense(4096,activation='relu',name="fc7",trainable=True)(x)

x = keras.layers.Dense(4096,activation='relu',name="fc8",trainable=True)(x)

x = keras.layers.Dense(288,activation='linear',name="fc10",trainable=True)(x)

The fact is that,the outputs are not really of size 288, they have all variable lenght, but I zero-padded all of them in order to have a fixed-size output.

In the other hand, I also trained a neural net that predicts the lenght of these outputs giving the same 3D input and it does it pretty well.

What I want is to use the output of this second network to crop the output of my first network . For example ,if my second network predicted a value of 285,the first network will crop the output from 288 to 285.

Is it possible on keras ? Or could you give me and advice of how to do this task(Predict the good coordinates with the correct length)?

Thank you very much