3jKiSSKJr SSKJrJr SSKJr SSKrS"S#SjjrS"S#Sjjr S"S#Sjjr "S S \5r S$S%S jjr S$S&S jjr "S S\RR5r"SS\5r"SS\5r"SS\5r"SS\5r"SS\5r"SS\5r\ R.SS4S'Sjjr\ R.SS4S'Sjjr\ R.SS4S'Sjjr\ R.SS4S'Sjjr\ R.SS4S'S jjr\ R.SS4S'S!jjrg)() annotations)Enummember)partialN?c6UR5RSU- 5nUSU-S- -S-nUSU-S- -S-n[R"US:U[R"US:*U[R"SUR UR S 955$) aImplementation of a 2nd-order polynomial kernel for Kernel density estimate (Xu et al., 2008). Support: [-window_radius, window_radius]. Returns 0 outside this range. Ref: "Parzen-Window Based Normalized Mutual Information for Medical Image Registration", Eq. 22. Args: x (torch.Tensor): signal, any shape window_radius (float): radius of window for the kernel Returns: torch.Tensor: transformed signal rgg?g?g @gffffff??)devicedtype)absmultorchwheretensorr r )x window_radiusx_abspoly1poly2s Z/home/wildlama/miniconda3/lib/python3.13/site-packages/kornia/losses/mutual_information.py xu_kernelrs EEGKKm+ ,E TE\C' (3 .E S5[3& '# -E ;;  UEKK eU\\#VWV^V^fgfmfm=no cb[R"U5n[R"X:*SS5$)aImplementation of a rectangular kernel. Support: [-window_radius, window_radius]. Returns 1.0 inside this range, 0.0 otherwise. Args: x (torch.Tensor): signal, any shape window_radius (float): radius of window for the kernel Returns: torch.Tensor: transformed signal rr )rr r)rrs rrectangular_kernelr0s'  ! A ;;q)3 44rcUn[R"U5U:*n[R"SX- S--5US[R-S--- n[R"X4S5$)a[Implementation of a truncated Gaussian kernel. Support: [-window_radius, window_radius]. Returns Gaussian value inside this range, 0.0 otherwise. Sigma is set to window_radius. Args: x (torch.Tensor): signal, any shape window_radius (float): radius of window for the kernel (used as sigma) Returns: torch.Tensor: transformed signal gr r )rr exppir)rrsigmamask gaussian_vals rtruncated_gaussian_kernelr#@s^ E 99Q<= (D99TQY1$445!ehh,SVAV9VWL ;;t3 //rcD\rSrSr\"\5r\"\5r\"\ 5r Sr g)MIKernelUN) __name__ __module__ __qualname____firstlineno__rrxur rectangularr#truncated_gaussian__static_attributes__r'rrr%r%Us#  B+,K 9:rr%cURSS9up4URSS9upTXS- RS5n[R"Xb:XRS5- U- U-S5$)Ndimr)minmax unsqueezerr)datanum_binsepsmin_val_max_valdiffs r_normalize_signalr>[sgb!JGb!JG   ( ( ,D ;;tzD+<+)r 2C ))SD599S>)r 2C T>rct^\rSrSrSr\R SS4S U4SjjjrS SjrS Sjr Sr U=r $) EntropyBasedLossBasesacA base class for entropy-based loss functions using kernel density estimation. This class provides the foundation for computing entropy-based losses between signals by estimating probability distributions using kernel density estimation (KDE). It computes joint histograms and derives entropy measures that can be used to quantify the similarity or dissimilarity between signals. The class pre-processes a reference signal and provides methods to compute joint histograms with other signals, from which various entropy measures (joint, marginal, conditional, mutual information) can be derived in subclasses. @rc>[TU]5 [R"UR5R UlUR S[XUR 55 X0lU[;a[S[[5S35e[URUS9UlX@l[R "URUR"R$S9Ulg)a,Initialize the entropy-based loss base module. Args: reference_signal (torch.Tensor): reference signal to which other signals will be compared by the forward method kernel_function (MIKernel): Used kernel function for kernel density estimate, by default MIKernel.xu num_bins (int): number of signal value bins in kernel density estimate, by default 64 window_radius (float): radius of the kernel's support interval, by default 1.0 Raises: ValueError: If kernel_function is not a valid MIKernel member. signalzRThe passed_kernel_function is not an accepted MIKernel, the available options are .)r)r N)super__init__rfinfor r9register_bufferr>r8r% ValueErrorlistrvaluekernel_functionrarangerRr bin_centersselfreference_signalr[r8r __class__s rrUEntropyBasedLossBase.__init__s, ;;/556:: X'89IUYU]U]'^_ ( *deijresdttuv  ''<'P>PQrcURURR:wa0[SURSURRS35e[XRUS9nUR R S5URR S5- nUR R S5UR S5- nURU5nURU5n[R"SXV5nU$)aCompute the joint histogram between the reference signal and another signal. Uses kernel density estimation to estimate the joint probability distribution between the reference signal and the provided other signal. The histogram is computed by evaluating kernel functions at discretized signal values. Args: other_signal (torch.Tensor): Signal to compare with the reference signal. Must have the same shape as the reference signal. eps (float): Epsilon value for numerical stability in computations. Returns: torch.Tensor: Joint histogram tensor with shape [..., num_bins, num_bins] representing the estimated joint probability distribution. Raises: ValueError: If other_signal has incompatible shape with reference signal. z*The two signals have incompatible shapes: z and rS)r8r9r1r@z...in,...jn->...ij) shaperRrXr>r8r]r6r[reinsum)r_ other_signalr9diff_1diff_2vals_1vals_2rEs r_compute_joint_histogram-EntropyBasedLossBase._compute_joint_histograms&   !2!2 2I,J\J\I]]bcgcncnctctbuuvwx x( SVW !!++B/$++2G2G2KK!!++B/,2H2H2LL%%f-%%f-,,';VLrc^URXR5n[X RS9$)aCompute entropy measures between the reference signal and another signal. Calculates joint entropy and marginal entropies based on the joint histogram of the two signals. Args: other_signal (torch.Tensor): Signal to compare with the reference signal. Must have the same shape as the reference_signal passed at instantiation. Returns: tuple[torch.Tensor, torch.Tensor, torch.Tensor]: A tuple containing: - Marginal entropy H(X) of reference signal - Marginal entropy H(Y) of other signal - Joint entropy H(X,Y) All tensors have the same batch dimensions as the input signals. Note: Subclasses should implement specific loss functions based on these entropy measures (e.g., mutual information). )r9)rkr9rL)r_rfrEs r entropiesEntropyBasedLossBase.entropiess'*77 hhO,_((KKr)r]r9r[r8r r` torch.Tensorr[r%r8intrfloatreturnNone)rfrqr9rsrtrq)rfrqrtz/tuple[torch.Tensor, torch.Tensor, torch.Tensor]) r(r)r*r+__doc__r%r,rUrkrnr/ __classcell__ras@rrNrNssk %-KK" R& R" R R  R  R RDBLLrrNc\rSrSrSSjrSrg) MILossFromRefc>URU5up#nX#-U- nU*$)aCompute differentiable mutual information for self.signal and other_signal. mi = (H(X) + H(Y) - H(X,Y)) To have a loss function, the opposite is returned. Can also handle two batches of flat tensors, then a batch of loss values is returned. Args: other_signal: Batch of flat tensors shape (B,N) where B is the tuple of batch dimensions for self.signal, possibly empty. Returns: torch.Tensor: tensor of losses, shape B as above rn)r_rfrJrKrImis rforwardMILossFromRef.forwards) 5$ Y s rr'Nrfrqrtrqr(r)r*r+rr/r'rrrzrzsrrzc\rSrSrSSjrSrg)NMILossFromRefc>URU5up#nX#-U- nU*$)aCompute differentiable normalized mutual information for self.signal and other_signal. nmi = (H(X) + H(Y)) / H(X,Y) To have a loss function, the opposite is returned. Can also handle two batches of flat tensors, then a batch of loss values is returned. Args: other_signal: Batch of flat tensors shape (B,N) where B is the tuple of batch dimensions for self.signal, possibly empty. Returns: torch.Tensor: tensor of losses, shape B as above r})r_rfrJrKrInmis rrNMILossFromRef.forwards) 5$yD t rr'Nrrr'rrrrsrrc^\rSrSrSr\R SS4S U4Sjjjr\S Sj5r S U4Sjjr Sr U=r $) MILossFromRef2Dia`Mutual Information loss module specifically designed for 2D image data. This class extends MILossFromRef to handle 2D image inputs by automatically reshaping batched 2D images into the format expected by the base entropy computation methods. It computes mutual information between reference 2D images and other images using kernel density estimation. rPcF>[TU]URU5X#U5 g)a8Initialize the 2D Mutual Information loss module. Args: reference_signal (torch.Tensor): reference signal to which other signals will be compared by the forward method. batch of 2D images, shape (B,H,W) where B are the batch dimensions. kernel_function (MIKernel): Used kernel function for kernel density estimate, by default MIKernel.xu num_bins (int): number of signal value bins in kernel density estimate, by default 64 window_radius (float): radius of the kernel's support interval, by default 1.0 NrTrU arrange_shaper^s rrUMILossFromRef2D.__init__#* ++,<=ZghrcDURURSSS-5$Nr@r1reshaperdrs rrMILossFromRef2D.arrange_shape&!~~fll3B/%788rc@>[TU]URU55$)aCompute differentiable mutual information for reference_signal and other_signal (both supposed 2D). mi = (H(X) + H(Y) - H(X,Y)) To have a loss function, the opposite is returned. Can also handle two batches of flat tensors, then a batch of loss values is returned. Args: other_signal: Batch of flat tensors same shape (B,H,W) as reference_signal passed for instantiation Returns: torch.Tensor: tensor of losses, shape B as above rTrrr_rfras rrMILossFromRef2D.forward*wt11,?@@rr'rprrqrtrqr r(r)r*r+rvr%r,rU staticmethodrrr/rwrxs@rrrx%-KK i&i"i i  i  ii.99AArrc^\rSrSrSr\R SS4S U4Sjjjr\S Sj5r S U4Sjjr Sr U=r $) MILossFromRef3Di;akMutual Information loss module specifically designed for 3D image data. This class extends MILossFromRef to handle 3D image inputs by automatically reshaping batched 3D images into the format expected by the base entropy computation methods. It computes normalized mutual information between reference 3D images and other images using kernel density estimation. rPrcF>[TU]URU5X#U5 g)a:Initialize the 3D Mutual Information loss module. Args: reference_signal (torch.Tensor): reference signal to which other signals will be compared by the forward method. batch of 3D images, shape (B,D,H,W) where B are the batch dimensions. kernel_function (MIKernel): Used kernel function for kernel density estimate, by default MIKernel.xu num_bins (int): number of signal value bins in kernel density estimate, by default 64 window_radius (float): radius of the kernel's support interval, by default 1.0 Nrr^s rrUMILossFromRef3D.__init__DrrcDURURSSS-5$Nrrrs rrMILossFromRef3D.arrange_shape[rrc@>[TU]URU55$)aCompute differentiable mutual information for reference_signal and other_signal (both supposed 3D). mi = (H(X) + H(Y) - H(X,Y)) To have a loss function, the opposite is returned. Can also handle two batches of flat tensors, then a batch of loss values is returned. Args: other_signal: Batch of flat tensors same shape (B,D,H,W) as reference_signal passed for instantiation Returns: torch.Tensor: tensor of losses, shape B as above rrs rrMILossFromRef3D.forward_rrr'rprrrrxs@rrr;rrrc^\rSrSrSr\R SS4S U4Sjjjr\S Sj5r S U4Sjjr Sr U=r $) NMILossFromRef2DipawNormalized Mutual Information loss module specifically designed for 2D image data. This class extends NMILossFromRef to handle 2D image inputs by automatically reshaping batched 2D images into the format expected by the base entropy computation methods. It computes normalized mutual information between reference 2D images and other images using kernel density estimation. rPrcF>[TU]URU5X#U5 g)aCInitialize the 2D Normalized Mutual Information loss module. Args: reference_signal (torch.Tensor): reference signal to which other signals will be compared by the forward method. batch of 2D images, shape (B,H,W) where B are the batch dimensions. kernel_function (MIKernel): Used kernel function for kernel density estimate, by default MIKernel.xu num_bins (int): number of signal value bins in kernel density estimate, by default 64 window_radius (float): radius of the kernel's support interval, by default 1.0 Nrr^s rrUNMILossFromRef2D.__init__yrrcDURURSSS-5$rrrs rrNMILossFromRef2D.arrange_shaperrc@>[TU]URU55$)aCompute differentiable mutual information for reference_signal and other_signal (both supposed 2D). nmi = (H(X) + H(Y)) / H(X,Y) To have a loss function, the opposite is returned. Can also handle two batches of tensors, then a batch of loss values is returned. Args: other_signal: Batch of tensors same shape (B,H,W) as reference_signal passed for instantiation Returns: torch.Tensor: tensor of losses, shape B as above rrs rrNMILossFromRef2D.forwardrrr'rprrrrxs@rrrprrrc^\rSrSrSr\R SS4S U4Sjjjr\S Sj5r S U4Sjjr Sr U=r $) NMILossFromRef3DialNormalized Mutual Information loss module specifically designed for 3D image data. This class extends NMILossFromRef to handle 3D image inputs by automatically reshaping batched 3D images into the format expected by the base entropy computation methods. It computes mutual information between reference 3D images and other images using kernel density estimation. rPrcF>[TU]URU5X#U5 g)aEInitialize the 3D Normalized Mutual Information loss module. Args: reference_signal (torch.Tensor): reference signal to which other signals will be compared by the forward method. batch of 3D images, shape (B,D,H,W) where B are the batch dimensions. kernel_function (MIKernel): Used kernel function for kernel density estimate, by default MIKernel.xu num_bins (int): number of signal value bins in kernel density estimate, by default 64 window_radius (float): radius of the kernel's support interval, by default 1.0 Nrr^s rrUNMILossFromRef3D.__init__rrcDURURSSS-5$rrrs rrNMILossFromRef3D.arrange_shaperrc@>[TU]URU55$)aCompute differentiable mutual information for reference_signal and other_signal (both supposed 3D). nmi = (H(X) + H(Y)) / H(X,Y) To have a loss function, the opposite is returned. Can also handle two batches of flat tensors, then a batch of loss values is returned. Args: other_signal: Batch of tensors, same shape (B,D,H,W) as reference_signal passed for instantiation Returns: torch.Tensor: tensor of losses, shape B as above rrs rrNMILossFromRef3D.forwardrrr'rprrrrxs@rrrrrrrPc8[XX4S9nURU5$)aDCompute differentiable mutual information for flat tensors. mi = (H(X) + H(Y) - H(X,Y)) To have a loss function, the opposite is returned. Can also handle two batches of flat tensors, then a batch of loss values is returned. Args: input (torch.Tensor): Batch of flat tensors shape (B,N) where B is any batch dimensions tuple, possibly empty. target (torch.Tensor): Batch of flat tensors, same shape as input. kernel_function (MIKernel): Used kernel function for kernel density estimate, by default MIKernel.xu num_bins (int): The number of bins used for KDE, defaults to 64. window_radius (float): The smoothing window radius in KDE, in terms of bin width units, defaults to 1. Returns: torch.Tensor: tensor of losses, shape B (common batch dims tuple of input and target) r`r[r8r)rzrinputtargetr[r8rmodules rmutual_information_lossrs$88F >>%  rc8[XX4S9nURU5$)a7Compute differentiable mutual information for 2d tensors. nmi = (H(X) + H(Y)) / H(X,Y) To have a loss function, the opposite is returned. Can also handle two batches of 2d tensors, then a batch of loss values is returned. Args: input (torch.Tensor): Batch of 2d tensors shape (B,H,W) where B is any batch dimensions tuple, possibly empty. target (torch.Tensor): Batch of 2d tensors, same shape as input. kernel_function (MIKernel): Used kernel function for kernel density estimate, by default MIKernel.xu num_bins (int): The number of bins used for KDE, defaults to 64. window_radius (float): The smoothing window radius in KDE, in terms of bin width units, defaults to 1. Returns: torch.Tensor: tensor of losses, shape B (common batch dims tuple of input and target) r)rrrs rmutual_information_loss_2dr$68F >>%  rc8[XX4S9nURU5$)a9Compute differentiable mutual information for 3d tensors. nmi = (H(X) + H(Y)) / H(X,Y) To have a loss function, the opposite is returned. Can also handle two batches of 3d tensors, then a batch of loss values is returned. Args: input (torch.Tensor): Batch of 3d tensors shape (B,D,H,W) where B is any batch dimensions tuple, possibly empty. target (torch.Tensor): Batch of 3d tensors, same shape as input. kernel_function (MIKernel): Used kernel function for kernel density estimate, by default MIKernel.xu num_bins (int): The number of bins used for KDE, defaults to 64. window_radius (float): The smoothing window radius in KDE, in terms of bin width units, defaults to 1. Returns: torch.Tensor: tensor of losses, shape B (common batch dims tuple of input and target) r)rrrs rmutual_information_loss_3drrrc<[UUUUS9nURU5$)aPCompute differentiable normalized mutual information for flat tensors. nmi = (H(X) + H(Y)) / H(X,Y) To have a loss function, the opposite is returned. Can also handle two batches of flat tensors, then a batch of loss values is returned. Args: input (torch.Tensor): Batch of flat tensors shape (B,N) where B is any batch dimensions tuple, possibly empty. target (torch.Tensor): Batch of flat tensors, same shape as input. kernel_function (MIKernel): Used kernel function for kernel density estimate, by default MIKernel.xu num_bins (int): The number of bins used for KDE, defaults to 64. window_radius (float): The smoothing window radius in KDE, in terms of bin width units, defaults to 1. Returns: torch.Tensor: tensor of losses, shape B (common batch dims tuple of input and target) r)rrrs r"normalized_mutual_information_lossr>s+8'# F >>%  rc8[XX4S9nURU5$)aACompute differentiable normalized mutual information for 2d tensors. mi = (H(X) + H(Y) - H(X,Y)) To have a loss function, the opposite is returned. Can also handle two batches of 2d tensors, then a batch of loss values is returned. Args: input (torch.Tensor): Batch of 2d tensors shape (B,H,W) where B is any batch dimensions tuple, possibly empty. target (torch.Tensor): Batch of 2d tensors, same shape as input. kernel_function (MIKernel): Used kernel function for kernel density estimate, by default MIKernel.xu num_bins (int): The number of bins used for KDE, defaults to 64. window_radius (float): The smoothing window radius in KDE, in terms of bin width units, defaults to 1. Returns: torch.Tensor: tensor of losses, shape B (common batch dims tuple of input and target) r)rrrs r%normalized_mutual_information_loss_2drc$68F >>%  rc8[XX4S9nURU5$)aCCompute differentiable normalized mutual information for 3d tensors. mi = (H(X) + H(Y) - H(X,Y)) To have a loss function, the opposite is returned. Can also handle two batches of 3d tensors, then a batch of loss values is returned. Args: input (torch.Tensor): Batch of 3d tensors shape (B,D,H,W) where B is any batch dimensions tuple, possibly empty. target (torch.Tensor): Batch of 3d tensors, same shape as input. kernel_function (MIKernel): Used kernel function for kernel density estimate, by default MIKernel.xu num_bins (int): The number of bins used for KDE, defaults to 64. window_radius (float): The smoothing window radius in KDE, in terms of bin width units, defaults to 1. Returns: torch.Tensor: tensor of losses, shape B (common batch dims tuple of input and target) r)rrrs r%normalized_mutual_information_loss_3drrr)r)rrqrrsrtrq)g:0yE>)r7rqr8rrr9rsrtrq)rErqr9rsrtrq) rrqrrqr[r%r8rrrrsrtrq) __future__renumrr functoolsrrrrr#r%r>rLnnModulerNrzrrrrrr,rrrrrrr'rrrsu"# . 5 0*;t; X  fL588??fLR(*)*2Am2Aj2Am2Aj2A~2Aj2A~2Ap!)  ! ! !! !  !  !J!)  ! ! !! !  !  !H!)  ! ! !! !  !  !H!)  "! "! "!"! "!  "!  "!P!)  ! ! !! !  !  !H!)  ! ! !! !  !  !r