3jW)SSKJr SSKJr SSKrSSKJr "SS\R 5r"SS\5r"S S \5r g) ) annotations)CallableN)nnch^\rSrSr%SrS\S'S\S'S S U4SjjjrS SjrSSjrSS jr S r U=r $)_HausdorffERLossBaseaBase class for binary Hausdorff loss based on morphological erosion. This is an Hausdorff Distance (HD) Loss that based on morphological erosion,which provided a differentiable approximation of Hausdorff distance as stated in :cite:`karimi2019reducing`. The code is refactored on top of `here `__. Args: alpha: controls the erosion rate in each iteration. k: the number of iterations of erosion. reduction: Specifies the reduction to apply to the output: 'none' | 'mean' | 'sum'. 'none': no reduction will be applied, 'mean': the weighted mean of the output is taken, 'sum': the output will be summed. Returns: Estimated Hausdorff Loss. zCallable[..., torch.Tensor]convmax_poolc>[TU]5 XlX lX0lUR SUR 55 g)Nkernel)super__init__alphak reductionregister_buffer get_kernel)selfrrr __class__s Q/home/wildlama/miniconda3/lib/python3.13/site-packages/kornia/losses/hausdorff.pyr_HausdorffERLossBase.__init__1s4  " Xt'89c[e),Get kernel for image morphology convolution.)NotImplementedError)rs rr_HausdorffERLossBase.get_kernel8s!!rcX- S-n[R"URURURS9n[R "X1RURS9n[R "X1R[RS9nURS5S- S-n[UR5HnURX4USS9n U S- n SXS:'URU 5n URU *5*n X- S:gn U R5nUR5(a"X- X- - n[R"Xm-X5n XZUS-UR ---nU nM U$)Ndevicedtype)weightpaddinggroupsg?r)torch as_tensorr r r! zeros_like ones_likeboolsizerangerr r squeezeanywherer)rpredtargetboundr erodedmaskr%rdilationerosion erosion_max erosion_min_to_normto_norm_erosion_to_fills rperform_erosion$_HausdorffERLossBase.perform_erosion<sI1$T[[ S!!% 4::Nu[[ K;;r?Q&1,tvvAyywqyQHnG#$GaK --0K=='22K#1a7H&&(G{{}}%,$9k>W#X ++do7GQQ4::(===FE-0 rcURSSURSS:Xa9URS5URS5:XaURS5S:Xd&[SURSURS35eURS5UR5R 5:a [S5e[ R "[URS55Vs/sHnURUSS2X3S-24[ R"X#:H[ R"SURURS 9[ R"SURURS 955PM sn5nURS :XaUR5nU$URS :XaUR5nU$URS :XaU$[!S URS35es snf)aCCompute Hausdorff loss. Args: pred: predicted torch.Tensor with a shape of :math:`(B, C, H, W)` or :math:`(B, C, D, H, W)`. Each channel is as binary as: 1 -> fg, 0 -> bg. target: target torch.Tensor with a shape of :math:`(B, 1, H, W)` or :math:`(B, C, D, H, W)`. Returns: Estimated Hausdorff Loss. rNrr#zTPrediction and target need to be of same size, and target should not be one-hot.Got z and .zInvalid target value.rmeansumnonez reduction `z` has not been implemented yet.)shaper, ValueErrormaxitemr'stackr-r=r0tensorr r!rrArBr)rr1r2iouts rforward_HausdorffERLossBase.forward_s 12&,,qr"22tyy|v{{ST~7UZ`ZeZefgZhlmZmzzl% ~Q8  99Q<&**,++- -45 5kktyy|, -A$$AAI&KK  Qv}}FLLQ Qv}}FLLQ-    >>V #((*C ^^u $'')C  ^^v %  & DNN3CCb&cd d) s(BG.)rrr)g@ rA)rfloatrintrstrreturnNonerR torch.Tensorr1rUr2rUrRrU) __name__ __module__ __qualname____firstlineno____doc____annotations__rrr=rL__static_attributes__ __classcell__rs@rrrs5& &%))::"!F,,rrcv^\rSrSrSr\R r\R"S5r SSjr SU4Sjjr Sr U=r$) HausdorffERLossaBinary Hausdorff loss based on morphological erosion. Hausdorff Distance loss measures the maximum distance of a predicted segmentation boundary to the nearest ground-truth edge pixel. For two segmentation point sets X and Y , the one-sided HD from X to Y is defined as: .. math:: hd(X,Y) = \max_{x \in X} \min_{y \in Y}||x - y||_2 Furthermore, the bidirectional HD is: .. math:: HD(X,Y) = max(hd(X, Y), hd(Y, X)) This is an Hausdorff Distance (HD) Loss that based on morphological erosion, which provided a differentiable approximation of Hausdorff distance as stated in :cite:`karimi2019reducing`. The code is refactored on top of `here `__. Args: alpha: controls the erosion rate in each iteration. k: the number of iterations of erosion. reduction: Specifies the reduction to apply to the output: 'none' | 'mean' | 'sum'. 'none': no reduction will be applied, 'mean': the weighted mean of the output is taken, 'sum': the output will be summed. Examples: >>> hdloss = HausdorffERLoss() >>> input = torch.randn(5, 3, 20, 20) >>> target = (torch.rand(5, 1, 20, 20) * 2).long() >>> res = hdloss(input, target) r#cV[R"/SQ/SQ/SQ//5nUS-nUS$)rrr#rr#r#r#g?N)r'rI)rcrossr s rrHausdorffERLoss.get_kernels- y)Y?@Ad|rc >UR5S:wa[SUR5S35eUR5URS5:a2UR 5S:aUR [ R:Xd@[SURS5SUR 5SUR5S 35e[TU]%X5$) a Compute Hausdorff loss. Args: pred: predicted torch.Tensor with a shape of :math:`(B, C, H, W)`. Each channel is as binary as: 1 -> fg, 0 -> bg. target: target torch.Tensor with a shape of :math:`(B, 1, H, W)`. Returns: Estimated Hausdorff Loss. zOnly 2D images supported. Got r@r#rz+Expect long type target value in range (0, z). (z, )) dimrErFr,minr!r'longr rLrr1r2rs rrLHausdorffERLoss.forwards 88:?=dhhj\KL L tyy|+ 0AfllV[V`V`F`=diil^4PVPZPZP\~]_`f`j`j`l_mmno wt,,rrTrV)rWrXrYrZr[r'conv2dr rAdaptiveMaxPool2dr rrLr]r^r_s@rraras3"H <`__. Args: alpha: controls the erosion rate in each iteration. k: the number of iterations of erosion. reduction: Specifies the reduction to apply to the output: 'none' | 'mean' | 'sum'. 'none': no reduction will be applied, 'mean': the weighted mean of the output is taken, 'sum': the output will be summed. Examples: >>> hdloss = HausdorffERLoss3D() >>> input = torch.randn(5, 3, 20, 20, 20) >>> target = (torch.rand(5, 1, 20, 20, 20) * 2).long() >>> res = hdloss(input, target) r#c[R"/SQ/SQ/SQ//5n[R"/SQ/SQ/SQ//5n[R"X!U/S5S-nUS$)rrdre)rrrr#g$I$I?N)r'rIrH)rrfr3r s rrHausdorffERLoss3D.get_kernelsX y)Y?@A y)Y?@AeE2A6%@d|rc>UR5S:wa[SUR5S35e[TU] X5$)aCompute 3D Hausdorff loss. Args: pred: predicted torch.Tensor with a shape of :math:`(B, C, D, H, W)`. Each channel is as binary as: 1 -> fg, 0 -> bg. target: target torch.Tensor with a shape of :math:`(B, 1, D, H, W)`. Returns: Estimated Hausdorff Loss. zOnly 3D images supported. Got r@)rkrEr rLrns rrLHausdorffERLoss3D.forwards< 88:?=dhhj\KL Lwt,,rrprTrV)rWrXrYrZr[r'conv3dr rAdaptiveMaxPool3dr rrLr]r^r_s@rrtrts3"H <rsE$# q299qhA-*A-HA-,A-r