SAMed
SAMed copied to clipboard
hitachinsk
semantic categories
hi! How do you achieve showing different semantic categories on one image? I don't seem to see it in the code.
I did not release the visualization codes. Since the semantic mask is a map with multiple numbers, you can assign each number a color as you want. And I think it's easy to implement.
@hitachinsk correct me if I'm wrong, but basically all that was changed was substituting the number of classes into the num_multimask_outputs value. The original SAM predicts three "multimask" outputs (I think sorted from high to low confidence).
To the best of my knowledge, the MLP hypernetworks predict separate masks, but is it reasonable to expect that the losses can actually propagate to the MLPs? What about another semantic segmentation case where instead of having each pixel on an image assigned to one category, you could have overlapping categories?
