How to use -on option in mvNCCompile

[idata]

There is a command line in doc: "mvNCCompile inception-v1.meta -s 12 -in=input -on=InceptionV1/Logits/Predictions/Reshape_1 -is 224 224 -o InceptionV1.graph"

I think we use -in=input because we have the following code:"image = tf.placeholder("float", [1, image_size, image_size, 3], name="input")"how do we come up -on= InceptionV1/Logits/Predictions/Reshape_1?Thanks.-Tony

[idata]

@tzxu Tensorflow is hierarchical in terms of how layers are defined. With the example you provided, "InceptionV1/Logits/Predictions/Reshape_1", the Reshape_1 layer is being used as the output node and it is located inside of the Predictions function which is inside of the Logits function which is in the InceptionV1 network.

In regards to finding the input and output nodes for your network, you can try using the summarize_graph tool found at https://github.com/tensorflow/tensorflow/tree/master/tensorflow/tools/graph_transforms#inspecting-graphs to inspect your model and it will try to find the input and output nodes for your model.

[idata]

Thanks.

[idata]

Here is what I have done to find out the nodes' names.

in the inception_v1.py: add one line for tensorboard to display

with tf.Session() as sess:

init_fn(sess) summary_writer = tf.summary.FileWriter("./tensorflow_logs/inception_v1/", graph=tf.get_default_graph()) #add a line here saver = tf.train.Saver(tf.global_variables()) saver.save(sess, "output/"+name)

then run:

1. python3 inception_v1.py
2. tensorboard --logdir=./tensorflow_logs
3. go to a browser http://localhost:6006
4. click on graph, in the graph, we can see all nodes' names.

[idata]

@tzxu Awesome! Thank you for posting that!