I'm trying to set CIFAR10's tf-slim model to have input of dynamic batch, height, width and single channel, i.e. monochromatic images of different sizes. Given that all shapes but channel size are dynamic, the output shape of tf.flatten is (?, ?). Is there any way to circumvent this? I'm trying to adapt CIFAR10 to tf's DeepDream tutorial that uses InceptionV3 with an unspecific input shape.
I'm assuming this happens because CIFAR10 is not fully convolutional
links: "{{ [] if var_db_link == '' else var_db_link }}"
ValueError: The last dimension of the inputs to global should be defined. Found F1::alpha(10,10)
F2::alpha(20,30)
F3::beta()
alpha(x,y)
{
myvar := x*2 + y
}
beta()
{
myvar := 47
}
.