lose function for semantic segmentation
I'm looking for a net for semantic segmentation.
I don't care much on the accuracy of the segmentation but I cant miss objects.
For example in the case of two obj one is much bigger the the other, I rather detect both of them than detect the big one perfectly and completely miss the small one.
I can't find a loss function that will take into consideration the detection of obj, all the loss function that I found just compare pixel by pixel.
Any idea what I can do? loss function that will help?