I am implementing generalized dice loss python layer,
I have extracted the labels to be treated as binary segmentation problem, Now I have some misunderstanding on background labels, which is 0 values and is the feature map[0] in a subvolume NxCxDxWxH (C stands for class numbers: 5 classes, and D stands for the depth of subvolume).
according to this line for binary segmentation with Blob shape with 2 feature maps, it is clear how to do that, what about generalized dice loss? Because I have 4 extra feature maps for foreground classes. How can I calculate the bottom[0].diff[i,0,:] for the background here?
Hi,
I am implementing generalized dice loss python layer, I have extracted the labels to be treated as binary segmentation problem, Now I have some misunderstanding on background labels, which is
0values and is the feature map[0] in a subvolume NxCxDxWxH (C stands for class numbers: 5 classes, and D stands for the depth of subvolume).according to this line for binary segmentation with Blob shape with 2 feature maps, it is clear how to do that, what about generalized dice loss? Because I have 4 extra feature maps for foreground classes. How can I calculate the
bottom[0].diff[i,0,:]for the background here?Your expert opinion is really appreciated Thanks