I noticed that the appendix mentions you also experimented with semantic segmentation tasks. However, since semantic segmentation involves multiple classes, I’m curious about how a three-channel RGB label can be used to represent so many categories.
Based on the description in the paper, I assume that each class is assigned a specific color, and during the inference phase, the predicted RGB values are directly used to determine the class. Could you please confirm if this understanding is correct?
I noticed that the appendix mentions you also experimented with semantic segmentation tasks. However, since semantic segmentation involves multiple classes, I’m curious about how a three-channel RGB label can be used to represent so many categories.
Based on the description in the paper, I assume that each class is assigned a specific color, and during the inference phase, the predicted RGB values are directly used to determine the class. Could you please confirm if this understanding is correct?