keras_to_tensorflow icon indicating copy to clipboard operation
keras_to_tensorflow copied to clipboard

Is .pb going to be faster than .h5?

Open harryxu-yscz opened this issue 6 years ago • 1 comments

Hi all, based on your experiences, is the converted pb model going to be faster than the h5 model during inference? I tested on my CPU and it shows pb model is about 20% slower than the h5 model.

harryxu-yscz avatar Jul 20 '18 17:07 harryxu-yscz

This is just converting the keras model to tensorflow with no substantial change to the structure. However, given that redundant nodes are being pruned, I expect the tensorflow model to be faster. I am surprised that was not the case.

amir-abdi avatar Oct 16 '18 21:10 amir-abdi