Tensorflow Inception - the number of classes

Question

I am interested to know about the number of classes in Tensorflow Inception .

In their training script, they download a training set consisting of images and tags. Then, to calculate the loss, they determine the number of classes as:

# Number of classes in the Dataset label set plus 1. # Label 0 is reserved for an (unused) background class. num_classes = dataset.num_classes() + 1 

You can see that they use the "unused background class". You can also see this approach when creating your training set: build_image_data.py

So why do you need such an unused background class? (Moreover, you get one additional, but useless prediction from the output level)

Answer 1

This is the convention we use for all of our image datasets, and it seems that it wasn’t worth breaking it for this particular model. As an aside, I would like all academic classification datasets to have “none of the above” classes in their test scores. A classifier that does not know when it does not know is not so useful in practice. (Vanhoucke)

https://groups.google.com/a/tensorflow.org/forum/#!topic/discuss/9G-c2K_GCmk