Hi, for multiclass mask rcnn we should have both classes in one mask with two different labels? so can we get different instances of each class with different colors finally?
Fafa,
You may need to decide on the structure you want to use for your classes. So, if you want to consider a two-class problem, you may try every class containing different color instances. Not sure how would that help, though. From my experience, deep NN learning is color agnostic, and it focuses more on structure and shape. So it might work.
I hope this helps, and that I am not drifting away as the original post is a bit old and I am not in the area anymore, just some memory traces.
Thanks Deeply! I want to run multiclass Mask RCNN wich even distinguishes each class instance!
well, if by colors you mean labels, then yes. But they are not colors, as in color perception.
Hence, the target is a label map that we display in color for illustration purposes, no more.