What would be the output node of the InceptionV4 model and why?

Autonomous Machines Jetson & Embedded Systems Jetson TX2
1

Hi

I want to do a post training quantization. I have used Necron to visualize the model InceptionV4. I am not sure of the output node:

a) BiasAdd: InceptioV4/Logits/Logits/BiasAdd

b) FakeQuantWithMinMaxVars: InceptionV4/Logits/Logits/act_quant/FakeQuantWithMinMaxVar

c) Softmax: InceptionV4/Logits/Predictions

I do not know how to insert images here. But I have used Netron to visualize the model. And I have just published the same question in Stackoverflow: https://stackoverflow.com/questions/60640994/what-would-be-the-output-node-of-the-inceptionv4-model-and-why

2

Hi,

It should be the softmax layer.

Suppose this is a classification network, softmax can give you the approximate probability output.
And based on the graph, softmax is the output node.

Thanks.

3

Ok.

But in this webpage: https://github.com/NVIDIA-AI-IOT/tf_to_trt_image_classification?nvid=nv-int-jnwrtwtttwhjn-33356 it seems that

InceptionV4/Logits/Logits/BiasAdd

has been used.

I am confused now.

4

Hi,

The ‘softmax’ is the operation of final layer.
And ‘InceptionV4/Logits/Logits/BiasAdd’ is the output layer name.

Thanks.