dice about topointeraction HOT 4 CLOSED

Hello, the current code uses CrossEntropy with the critical_voxels_map.

If you want to use Dice loss instead, you can modify in the following way:

1. Replace CrossEntropy with Dice here:

   TopoInteraction/TI_Loss.py

   Line 25 in 1bcb467

   self.ce_loss_func = torch.nn.CrossEntropyLoss(reduction='none')

2. Also replace the usage of CrossEntropy with Dice here:

   TopoInteraction/TI_Loss.py

   Line 137 in 1bcb467

   ce_tensor = torch.unsqueeze(self.ce_loss_func(x.double(),y[:,0].long()),dim=1)

from topointeraction.

Hello, the current code uses CrossEntropy with the critical_voxels_map.

If you want to use Dice loss instead, you can modify in the following way:

1. Replace CrossEntropy with Dice here:

   TopoInteraction/TI_Loss.py

   Line 25 in 1bcb467

   self.ce_loss_func = torch.nn.CrossEntropyLoss(reduction='none')

2. Also replace the usage of CrossEntropy with Dice here:

   TopoInteraction/TI_Loss.py

   Line 137 in 1bcb467

   ce_tensor = torch.unsqueeze(self.ce_loss_func(x.double(),y[:,0].long()),dim=1)

Thank you for your answer.Yes I understand, but doesn't diceloss calculate a scalar? How does this apply to critical_voxels_map? Does pixel-by-pixel computing diceloss yield vectors of the same size 。

from topointeraction.

The logic is as follows:

1. Compute the critical_voxels_map
2. A = Multiply the prediction with critical_voxels_map
3. B = Multiply the GT with critical_voxels_map
4. Compute Dice score between A and B

from topointeraction.

The logic is as follows:

1. Compute the critical_voxels_map
2. A = Multiply the prediction with critical_voxels_map
3. B = Multiply the GT with critical_voxels_map
4. Compute Dice score between A and B
   Thank you very much for your patience in answering my confusion, I benefited a lot!

from topointeraction.