3jSrSSKJr SSKJr SSKrSSKJs Jr SSKJr SSK J r J r SSK Jr SSKJr S S S jjr"S S \R$5rg)u6Module containing the Sørensen-Dice Coefficient loss.) annotations)OptionalN)nn) KORNIA_CHECKKORNIA_CHECK_IS_TENSOR)mask_ignore_pixels)one_hotc`[U5 [UR5S:Xd[SUR35eURSSURSS:Xd%[SURSUR35eURUR:Xd%[SURSUR35eURSnS S 1n[ X';S US U35 [ R"USS 9n[X5up[XRSURURS9n U bU RS5 X-n X-nUb[US5 [ URSU:H=(a UR5U:HSUSUR35 [ URUR:HSURSUR35 OURURS5nSn US :Xa!S/U Q7n URSSS5nX-nX-n [R "X-U 5n [R "X-U 5n SU -X-- nU*S-nUS :Xa$X-R!S5UR!5- n[R""U5nU$)uCriterion that computes Sørensen-Dice Coefficient loss. According to [1], we compute the Sørensen-Dice Coefficient as follows: .. math:: \text{Dice}(x, class) = \frac{2 |X \cap Y|}{|X| + |Y|} Where: - :math:`X` expects to be the scores of each class. - :math:`Y` expects to be the one-hot torch.Tensor with the class labels. the loss, is finally computed as: .. math:: \text{loss}(x, class) = 1 - \text{Dice}(x, class) Reference: [1] https://en.wikipedia.org/wiki/S%C3%B8rensen%E2%80%93Dice_coefficient Args: pred: logits torch.Tensor with shape :math:`(N, C, H, W)` where C = number of classes. target: labels torch.Tensor with shape :math:`(N, H, W)` where each value is in range :math:`0 ≤ targets[i] ≤ C-1`. average: Reduction applied in multi-class scenario: - ``'micro'`` [default]: Calculate the loss across all classes. - ``'macro'``: Calculate the loss for each class separately and average the metrics across classes. eps: Scalar to enforce numerical stabiliy. weight: weights for classes with shape :math:`(num\_of\_classes,)`. ignore_index: labels with this value are ignored in the loss computation. Return: One-element torch.Tensor of the computed loss. Example: >>> N = 5 # num_classes >>> pred = torch.randn(1, N, 3, 5, requires_grad=True) >>> target = torch.empty(1, 3, 5, dtype=torch.long).random_(N) >>> output = dice_loss(pred, target) >>> output.backward() z,Invalid pred shape, we expect BxNxHxW. Got: Nz.pred and target shapes must be the same. Got: z and z1pred and target must be in the same device. Got: micromacrozThe `average` has to be one of z. Got: )dim) num_classesdevicedtypez$weight must be torch.Tensor or None.rz)weight shape must be (num_of_classes,): (z,), got z1weight and pred must be in the same device. Got: )g@g?)rlenshape ValueErrorrrFsoftmaxrr r unsqueeze_numelnew_onesviewtorchsummean)predtargetaverageepsweight ignore_indexnum_of_classespossible_average pred_soft target_masktarget_one_hotdims intersection cardinality dice_score dice_losss L/home/wildlama/miniconda3/lib/python3.13/site-packages/kornia/losses/dice.pyr2r2%sh4 tzz?a G |TUU ::bc?fll23/ /I$**UZ[a[g[gZhijj ;;&-- 'LT[[MY^_e_l_l^mnooZZ]N),0OP`Oaahiphq.rs ii!4I,VBF$+6zz!}UYU`U`hlhrhr#sNq!'5+ v'MN \\!_ . S6<<>^3S77GxPVP\P\~ ^   MMT[[ (? eTXT_T_S` a tzz!}-#D'zDzRA&& '099Y7>L))I6=K|#{'89J c!I'',,R06::<?  9%I cX^\rSrSrSrSSU4SjjjrSSjrSrU=r$) DiceLossuCriterion that computes Sørensen-Dice Coefficient loss. Dice-based objectives are common in medical and semantic segmentation because pixel classes are often highly imbalanced. This loss directly optimizes region-level agreement. Args: average: Reduction strategy for multi-class computation. Use "micro" to aggregate classes globally, or "macro" to average class-wise Dice scores. eps: Small constant added to the denominator for numerical stability. weight: Optional class-weight tensor of shape :math:`(C,)`. ignore_index: Label value to exclude from loss computation. Shapes: - pred: :math:`(N, C, H, W)` where C is the number of classes. - target: :math:`(N, H, W)` where each value is in the range :math:`[0, C-1]`. - Output: scalar by default. cR>[TU]5 XlX lX0lX@lg)aInitialize the Dice loss module. Args: average: Reduction strategy for multi-class computation. eps: Small constant added for numerical stability. weight: Optional class-weight tensor of shape :math:`(C,)`. ignore_index: Label value to exclude from loss computation. N)super__init__r%r&r'r()selfr%r&r'r( __class__s r3r:DiceLoss.__init__s%   (r4cp[XURURURUR5$)uCompute Dice loss from logits and integer labels. Args: pred: The input prediction logits with shape :math:`(N, C, H, W)`. target: The ground truth labels with shape :math:`(N, H, W)`. Returns: The computed Sørensen-Dice loss value. )r2r%r&r'r()r;r#r$s r3forwardDiceLoss.forwards)t||TXXt{{DL]L]^^r4)r%r&r(r'rg:0yE>Ni) r%strr&floatr'Optional[torch.Tensor]r( Optional[int]returnNone)r# torch.Tensorr$rHrFrH) __name__ __module__ __qualname____firstlineno____doc__r:r?__static_attributes__ __classcell__)r<s@r3r6r6s\()-&* )))' ) $ )  ))* _ _r4r6rA)r#rHr$rHr%rBr&rCr'rDr(rErFrH)rM __future__rtypingrr torch.nn.functionalr functionalrkornia.core.checkrrkornia.losses._utilsrkornia.losses.one_hotr r2Moduler6r4r3rYs$=" B3)%)"& t t tt  t # t  ttn2_ryy2_r4