My frozen graph in tensorflow is throwing an error if I use a different image resolution than the one used for training at the "tf.reshape" command.
My network is fully convolutional and before having to use the tf.reshape command I was able to use different resolution input images in my code.
Now I have a function in my code that reshapes the tensor (it just does a shuffle like in Shufflenet2) and this is throwing an error now. I have tried getting the shape dynamically using tf.shape and its still not working.
def channel_shuffle( x, num_groups=2,name='channelshuf',reuse=False):
with tf.variable_scope(name+'shuffle', reuse=reuse):
#n, h, w, c = x.shape.as_list()
n, h, w, c = tf.shape(x)[0],tf.shape(x)[1],tf.shape(x)[2],tf.shape(x)[3]
x_reshaped = tf.reshape(x, [n, h, w, num_groups, c // num_groups])
x_transposed = tf.transpose(x_reshaped, [0, 1, 2, 4, 3])
output = tf.reshape(x_transposed, [n, h, w, c])
return output
The error I am getting is this:
InvalidArgumentError (see above for traceback): Input to reshape is a tensor with 3340800 values, but the requested shape has 835200
thank you!
