#Class Num or #Class Num + 1

[ddz16]

Hello, I noticed that the output of the classification head in the code is #Class Num categories, but some code comments also mention #Class Num+1. So I have the following two questions:

1. Is it #Class Num or #Class Num+1?
2. If it is #Class Num, how do we differentiate the background class?

[happyharrycn]

The output dimension of the classification head is #Class, denoting #Class-way binary classification. Each dimension here represents the probability of one category. There is no need to encode the background class in this K-way classification setting.

I can't recall where this #Class Num+1 was mentioned in our comments. If you can pin down the code, we will take a look.

[ddz16]

Thanks for your reply. The #Class Num+1 comments are in the meta_archs.py: and

[tzzcl]

These are old and wrong comments, we will remove/correct these comments shortly. Thank you for pointing it out.