How to understand "reduce_zero_label"?

[WilliamJames0741]

I found that in the dataset "ade20k", the "reduce_zero_label" is set to "True", and the net like segnext is set to 150 classes(but include "background" there are 151 classes). However, when I use the well trained model to inference, its output has the "background" class. I wound why and how to understand the "reduce_zero_label" set? And how can I achieve the same functionality?

[zhouzaida]

Hi, you can continue the question in https://github.com/open-mmlab/mmsegmentation/discussions.

[WilliamJames0741]

谢谢，您的邮件我已收到！