Let's use TensorBoard to see what we can find. In the GRAPHS section of the TensorBoard view of our model, you can see, as shown in Figure 7.3, that the BiasAdd node, highlighted in red is the input of an ArgMax operation used to calculate accuracy, as well as the input of a softmax operation. We can use the SparseSoftmaxCrossEntropyWithLogits (shown in Figure 7.3 only as SparseSiftnaxCr...) operation or just dense/BiasAdd as the output node name, but we'll use ArgMax and dense/BiasAdd as the two output node names for the freeze_graph tool so we can more easily see the output of the final dense layer as well as the ArgMax result:
Figure 7.3: Showing possible output node names of the model
Run the ...
