o oij6@sddlmZmZmZddlZddlmZddlmZm Z m Z gdZ GdddeZ dd ed e d edededeeeeeff ddZdd ed e d ededef ddZdd ededed e def ddZdS))ListOptionalTupleN) ImageModule)Tensor concatenatetensor) ZCAWhiteninglinear_transformzca_mean zca_whitenc seZdZdZ     ddededed ed ed d f fd d Zded dfddZ ddeded efddZ ded efddZ Z S)r as Compute the ZCA whitening matrix transform and the mean vector and applies the transform to the data. The data tensor is flattened, and the mean :math:`\mathbf{\mu}` and covariance matrix :math:`\mathbf{\Sigma}` are computed from the flattened data :math:`\mathbf{X} \in \mathbb{R}^{N \times D}`, where :math:`N` is the sample size and :math:`D` is flattened dimensionality (e.g. for a tensor with size 5x3x2x2 :math:`N = 5` and :math:`D = 12`). The ZCA whitening transform is given by: .. math:: \mathbf{X}_{\text{zca}} = (\mathbf{X - \mu})(US^{-\frac{1}{2}}U^T)^T where :math:`U` are the eigenvectors of :math:`\Sigma` and :math:`S` contain the corresponding eigenvalues of :math:`\Sigma`. After the transform is applied, the output is reshaped to same shape. Args: dim: Determines the dimension that represents the samples axis. eps: a small number used for numerical stability. unbiased: Whether to use the biased estimate of the covariance matrix. compute_inv: Compute the inverse transform matrix. detach_transforms: Detaches gradient from the ZCA fitting. shape: - x: :math:`(D_0,...,D_{\text{dim}},...,D_N)` is a batch of N-D tensors. - x_whiten: :math:`(D_0,...,D_{\text{dim}},...,D_N)` same shape as input. .. note:: See a working example `here `__. Examples: >>> x = torch.tensor([[0,1],[1,0],[-1,0],[0,-1]], dtype = torch.float32) >>> zca = ZCAWhitening().fit(x) >>> x_whiten = zca(x) >>> zca = ZCAWhitening() >>> x_whiten = zca(x, include_fit = True) # Includes the fitting step >>> x_whiten = zca(x) # Can run now without the fitting set >>> # Enable backprop through ZCA fitting process >>> zca = ZCAWhitening(detach_transforms = False) >>> x_whiten = zca(x, include_fit = True) # Includes the fitting step Note: This implementation uses :py:meth:`~torch.svd` which yields NaNs in the backwards step if the singular values are not unique. See `here `_ for more information. References: [1] `Stanford PCA & ZCA whitening tutorial `_ rư>TFdimepsunbiaseddetach_transforms compute_invreturnNcs>t||_||_||_||_||_d|_|||dS)NF)super__init__rrrrrfitted)selfrrrrr __class__F/home/ubuntu/.local/lib/python3.10/site-packages/kornia/enhance/zca.pyrQs zZCAWhitening.__init__xcCs|t||j|j|j|j\}}}||_||_|dur!tdg|_ n||_ |j r9|j |_|j |_|j |_ d|_ |S)zFit ZCA whitening matrices to the data. Args: x: Input data. Returns: Returns a fitted ZCAWhiten object instance. NrT) r rrrr mean_vectortransform_matrixtorchempty transform_invrdetachr)rrTmeanT_invrrrfitgs    zZCAWhitening.fit include_fitcCs4|r|||jstdt||j|j|j}|S)zApply the whitening transform to the data. Args: x: Input data. include_fit: Indicates whether to fit the data as part of the forward pass. Returns: The transformed data. TNeeds to be fitted first before running. Please call fit or set include_fit to True.)r&r RuntimeErrorr rrr)rrr'x_whitenrrrforwards  zZCAWhitening.forwardcCsP|jstd|jstd|jdurtd|j|j }t||j|}|S)zApply the inverse transform to the whitened data. Args: x: Whitened data. Returns: Original data. r(z;Did not compute inverse ZCA. Please set compute_inv to TrueNz6The transform inverse should be a Tensor. Gotcha None.) rr)rr! TypeErrorrmmrr )rrmean_invyrrrinverse_transforms  zZCAWhitening.inverse_transform)rr TTF)F) __name__ __module__ __qualname____doc__intfloatboolrrr&r+r0 __classcell__rrrrr s.6r Tr Finprrrreturn_inversercCs6t|tstdt|t|tstdt|t|ts*tdt|t|ts8tdt|t|tsFtdt||}|t|ksW|t| krkt dt| dt|dd ||d krut||}t t d |t |dt|g}t t |g|g}||}||} t |d |||dd } tt | } t j|d d d } | d| f} || | f| } | | }|r|t| d}n|t| }t j|\}}}|dd}t ||}|||}d }|r|t |||}|| |fS)aUCompute the ZCA whitening matrix and mean vector. The output can be used with :py:meth:`~kornia.color.linear_transform`. See :class:`~kornia.color.ZCAWhitening` for details. Args: inp: input data tensor. dim: Specifies the dimension that serves as the samples dimension. unbiased: Whether to use the unbiased estimate of the covariance matrix. eps: a small number used for numerical stability. return_inverse: Whether to return the inverse ZCA transform. Shapes: - inp: :math:`(D_0,...,D_{\text{dim}},...,D_N)` is a batch of N-D tensors. - transform_matrix: :math:`(\Pi_{d=0,d\neq \text{dim}}^N D_d, \Pi_{d=0,d\neq \text{dim}}^N D_d)` - mean_vector: :math:`(1, \Pi_{d=0,d\neq \text{dim}}^N D_d)` - inv_transform: same shape as the transform matrix Returns: A tuple containing the ZCA matrix and the mean vector. If return_inverse is set to True, then it returns the inverse ZCA matrix, otherwise it returns None. .. note:: See a working example `here `__. Examples: >>> x = torch.tensor([[0,1],[1,0],[-1,0],[0,-1]], dtype = torch.float32) >>> transform_matrix, mean_vector,_ = zca_mean(x) # Returns transformation matrix and data mean >>> x = torch.rand(3,20,2,2) >>> transform_matrix, mean_vector, inv_transform = zca_mean(x, dim = 1, return_inverse = True) >>> # transform_matrix.size() equals (12,12) and the mean vector.size equal (1,12) Input type is not a Tensor. Got eps type is not a float. Gotunbiased type is not bool. Got(Argument 'dim' must be of type int. Got z-Argument return_inverse must be of type bool 4Dimension out of range (expected to be in range of [, ], but got rNT)rkeepdim) isinstancerr,typer6r7r5sizelen IndexErrorrrarangertolistpermuteproditemr$reshapetr-linalgsvdrsqrtsqrt)r9rrrr:inp_size feat_dims new_order inp_permuteN feature_sizes num_featuresr$inp_center_flatcovUS_ S_inv_rootr#r%rrrr sL %    " $     r cCst|tstdt|t|tstdt|t|ts*tdt|t|ts8tdt|t||||d\}}}t||||}|S)aApply ZCA whitening transform. See :class:`~kornia.color.ZCAWhitening` for details. Args: inp: input data tensor. dim: Specifies the dimension that serves as the samples dimension. unbiased: Whether to use the unbiased estimate of the covariance matrix. eps: a small number used for numerical stability. Returns: Whiten Input data. .. note:: See a working example `here `__. Examples: >>> x = torch.tensor([[0,1],[1,0],[-1,0]], dtype = torch.float32) >>> zca_whiten(x) tensor([[ 0.0000, 1.1547], [ 1.0000, -0.5773], [-1.0000, -0.5773]]) r;r<r=r>F) rErr,rFr6r7r5r r )r9rrr transformr$r` inp_whitenrrrr s    r rrcCs&|}|t|ks|t| kr%tdt| dt|dd||dkr/t||}ttd|t|dt|g}tt|g|g}t|}|}|} t|d|||dd} t t |  } | |} | d| f} | |}||}| | }| | }|S)a2Given a transformation matrix and a mean vector, this function will flatten the input tensor along the given dimension and subtract the mean vector from it. Then the dot product with the transformation matrix will be computed and then the resulting tensor is reshaped to the original input shape. .. math:: \mathbf{X}_{T} = (\mathbf{X - \mu})(T) Args: inp: Input data :math:`X`. transform_matrix: Transform matrix :math:`T`. mean_vector: mean vector :math:`\mu`. dim: Batch dimension. Shapes: - inp: :math:`(D_0,...,D_{\text{dim}},...,D_N)` is a batch of N-D tensors. - transform_matrix: :math:`(\Pi_{d=0,d\neq \text{dim}}^N D_d, \Pi_{d=0,d\neq \text{dim}}^N D_d)` - mean_vector: :math:`(1, \Pi_{d=0,d\neq \text{dim}}^N D_d)` Returns: Transformed data. Example: >>> # Example where dim = 3 >>> inp = torch.ones((10,3,4,5)) >>> transform_mat = torch.ones((10*3*4,10*3*4)) >>> mean = 2*torch.ones((1,10*3*4)) >>> out = linear_transform(inp, transform_mat, mean, 3) >>> print(out.shape, out.unique()) # Should a be (10,3,4,5) tensor of -120s torch.Size([10, 3, 4, 5]) tensor([-120.]) >>> # Example where dim = 0 >>> inp = torch.ones((10,2)) >>> transform_mat = torch.ones((2,2)) >>> mean = torch.zeros((1,2)) >>> out = linear_transform(inp, transform_mat, mean) >>> print(out.shape, out.unique()) # Should a be (10,2) tensor of 2s torch.Size([10, 2]) tensor([2.]) r?r@rArBrNrD)rGrHrIrrrJrargsortrKr5rMrNrLrOr-)r9rrrrUrVpermperm_invrW inv_orderrZr[rXinp_flat inp_centerinp_transformedrrrr As*)" $     r )rTr F)rTr )r)typingrrrr kornia.corerModulerrr__all__r r5r7r6r r r rrrrs.  b$-