A Beginner Guide to Get the Output and Weights from an Existing TensorFlow Model

In order to get the output or weights in an existing tensorflow model, you can do by following steps:

Step 1: Load an existing model

First, we should load the existing tensorflow model in your application script. We can use saver.restore() to load.

Here is the tutorial:

After you have loaded the model, you can get its output and all weights in it.

In tensorflow, we can use graph.get_operation_by_name(variable_name).outputs[0] to get the variable.

Here variable_name is the name of variable in your loaded tensorflow model.

For example, you may create a input tensor in your model.

        with tf.name_scope('input'):
            self.input_x = tf.placeholder(tf.float32, [None, self.feature_dim, None, 1], name="input_x")

You have trained this model, saved it. After you loaded it in another application, you can get this input variable as follows:

input_x = graph.get_operation_by_name("input/input_x").outputs[0]

Here input/input_x the name of variable self.input_x.

In order to get the name of all variables in one model, you can read this tutorial:

After getting the variable of a model, we can get its value and print it.

Here is the example:

inputx = sess.run(input_x)
print(inputx)