o jq:@sddlmZmZmZmZddlZddlmmZ ddl m Z d,dej de dej fdd Z d-d eej d eeefdeeefdeeeefddf ddZdededej fddZ        d.dej dej dededeeefde ded ed!ed"edeej fd#d$ZGd%d&d&e Z  d/dej dej d'ej d!ed"edeej eej ej fff d(d)Z  d/dej dej d'ej d!ed"edeej eej ej fff d*d+ZdS)0)ListOptionalTupleUnionN)_Lossmeanx reductionreturncCs<|dkr|S|dkr|jddS|dkr|jddStd)zReduce input in batch dimension if needed. Args: x: Tensor with shape (N, *). reduction: Specifies the reduction type: ``'none'`` | ``'mean'`` | ``'sum'``. Default: ``'mean'`` nonerrdimsumz:Unknown reduction. Expected one of {'none', 'mean', 'sum'})rr ValueError)rr rE/home/kuhnn/.local/lib/python3.10/site-packages/TTS/tts/utils/ssim.py_reduce s  rrggtensors dim_range data_range size_rangec Cs |d}|D]}t|sJdt||j|jks)Jd|jd|j|durD||ksCJd|d|n&||d|d ||d|d ksjJd |d||d|d kr||dksJd |dd|n,|d|d kr|d|kr|d ksnJd |dd|d d||d|d kr|d|ksJd |dd|||d ksJd|d d|qdS)a7Check that input(-s) satisfies the requirements Args: tensors: Tensors to check dim_range: Allowed number of dimensions. (min, max) data_range: Allowed range of values in tensors. (min, max) size_range: Dimensions to include in size comparison. (start_dim, end_dim + 1) FNrzExpected torch.Tensor, got zExpected tensors to be on z, got z%Expected tensors with same size, got z and z9Expected tensors with same size at given dimensions, got z$Expected number of dimensions to be z,Expected number of dimensions to be between z*Expected values to be greater or equal to z(Expected values to be lower or equal to )torch is_tensortypedevicesizer minmax)rrrrrtrrr_validate_inputs*$.2.$ ,,r# kernel_sizesigmacCsdtj|tjd}||dd8}|d}|d|d d|d}||}|dS)zReturns 2D Gaussian kernel N(0,`sigma`^2) Args: size: Size of the kernel sigma: Std of the distribution Returns: gaussian_kernel: Tensor with shape (1, kernel_size, kernel_size) )dtyper@r)rarangefloat32 unsqueezeexpr)r$r%coordsgrrrgaussian_filterEs &  r/ ??FT{Gz?Q?y kernel_sigmafull downsamplek1k2c Cs|ddksJd|dt||gdd|fd|t|}|t|}tdtt|dd d } | dkrJ|rJtj|| d }tj|| d }t|| |dddd |} | d krct nt } | ||| || d \} }| d}|d}t||}t||}|r||gS|S)aInterface of Structural Similarity (SSIM) index. Inputs supposed to be in range ``[0, data_range]``. To match performance with skimage and tensorflow set ``'downsample' = True``. Args: x: An input tensor. Shape :math:`(N, C, H, W)` or :math:`(N, C, H, W, 2)`. y: A target tensor. Shape :math:`(N, C, H, W)` or :math:`(N, C, H, W, 2)`. kernel_size: The side-length of the sliding window used in comparison. Must be an odd value. kernel_sigma: Sigma of normal distribution. data_range: Maximum value range of images (usually 1.0 or 255). reduction: Specifies the reduction type: ``'none'`` | ``'mean'`` | ``'sum'``. Default:``'mean'`` full: Return cs map or not. downsample: Perform average pool before SSIM computation. Default: True k1: Algorithm parameter, K1 (small constant). k2: Algorithm parameter, K2 (small constant). Try a larger K2 constant (e.g. 0.4) if you get a negative or NaN results. Returns: Value of Structural Similarity (SSIM) index. In case of 5D input tensors, complex value is returned as a tensor of size 2. References: Wang, Z., Bovik, A. C., Sheikh, H. R., & Simoncelli, E. P. (2004). Image quality assessment: From error visibility to structural similarity. IEEE Transactions on Image Processing, 13, 600-612. https://ece.uwaterloo.ca/~z70wang/publications/ssim.pdf, DOI: `10.1109/TIP.2003.819861` r(rKernel size must be odd, got [])r)rrN)r$r>)rr5kernelr9r:)r#floatr!roundr rF avg_pool2dr/repeattor _ssim_per_channel_complex_ssim_per_channelrr)rr5r$r6rr r7r8r9r:frA_compute_ssim_per_channelssim_mapcs_mapssim_valcsrrrssimWs$)  " "    rPcs~eZdZdZgdZ       dd ed ed ed ededede eefddffdd Z de j de j de j fddZ ZS)SSIMLossap Creates a criterion that measures the structural similarity index error between each element in the input :math:`x` and target :math:`y`. To match performance with skimage and tensorflow set ``'downsample' = True``. The unreduced (i.e. with :attr:`reduction` set to ``'none'``) loss can be described as: .. math:: SSIM = \{ssim_1,\dots,ssim_{N \times C}\}\\ ssim_{l}(x, y) = \frac{(2 \mu_x \mu_y + c_1) (2 \sigma_{xy} + c_2)} {(\mu_x^2 +\mu_y^2 + c_1)(\sigma_x^2 +\sigma_y^2 + c_2)}, where :math:`N` is the batch size, `C` is the channel size. If :attr:`reduction` is not ``'none'`` (default ``'mean'``), then: .. math:: SSIMLoss(x, y) = \begin{cases} \operatorname{mean}(1 - SSIM), & \text{if reduction} = \text{'mean';}\\ \operatorname{sum}(1 - SSIM), & \text{if reduction} = \text{'sum'.} \end{cases} :math:`x` and :math:`y` are tensors of arbitrary shapes with a total of :math:`n` elements each. The sum operation still operates over all the elements, and divides by :math:`n`. The division by :math:`n` can be avoided if one sets ``reduction = 'sum'``. In case of 5D input tensors, complex value is returned as a tensor of size 2. Args: kernel_size: By default, the mean and covariance of a pixel is obtained by convolution with given filter_size. kernel_sigma: Standard deviation for Gaussian kernel. k1: Coefficient related to c1 in the above equation. k2: Coefficient related to c2 in the above equation. downsample: Perform average pool before SSIM computation. Default: True reduction: Specifies the reduction type: ``'none'`` | ``'mean'`` | ``'sum'``. Default:``'mean'`` data_range: Maximum value range of images (usually 1.0 or 255). Examples: >>> loss = SSIMLoss() >>> x = torch.rand(3, 3, 256, 256, requires_grad=True) >>> y = torch.rand(3, 3, 256, 256) >>> output = loss(x, y) >>> output.backward() References: Wang, Z., Bovik, A. C., Sheikh, H. R., & Simoncelli, E. P. (2004). Image quality assessment: From error visibility to structural similarity. IEEE Transactions on Image Processing, 13, 600-612. https://ece.uwaterloo.ca/~z70wang/publications/ssim.pdf, DOI:`10.1109/TIP.2003.819861` )r$r9r:r%rAr r0r1r3r4Trr2r$r6r9r:r8r rr NcsTt||_||_|ddksJd|d||_||_||_||_||_dS)Nr(rr;r<) super__init__r r$r6r9r:r8r)selfr$r6r9r:r8r r __class__rrrSs  zSSIMLoss.__init__rr5c Cs8t|||j|j|j|j|jd|j|jd }t ||S)aComputation of Structural Similarity (SSIM) index as a loss function. Args: x: An input tensor. Shape :math:`(N, C, H, W)` or :math:`(N, C, H, W, 2)`. y: A target tensor. Shape :math:`(N, C, H, W)` or :math:`(N, C, H, W, 2)`. Returns: Value of SSIM loss to be minimized, i.e ``1 - ssim`` in [0, 1] range. In case of 5D input tensors, complex value is returned as a tensor of size 2. F) rr5r$r6r8rr r7r9r:) rPr$r6r8rr r9r:r ones_like)rTrr5scorerrrforwards  zSSIMLoss.forward)r0r1r3r4Trr2)__name__ __module__ __qualname____doc__ __constants__intrBboolstrrrSrTensorrY __classcell__rrrUrrQs86  $rQrAcCsF|d|dks|d|dkr"td|d||d}|d}|d}tj||dd|d}tj||dd|d} |d} | d} || } tj|d|dd|d| } tj|d|dd|d| }tj|||dd|d| }d ||| ||}d | || | ||}|jd d }|jd d }||fS) aCalculate Structural Similarity (SSIM) index for X and Y per channel. Args: x: An input tensor. Shape :math:`(N, C, H, W)`. y: A target tensor. Shape :math:`(N, C, H, W)`. kernel: 2D Gaussian kernel. k1: Algorithm parameter, K1 (small constant, see [1]). k2: Algorithm parameter, K2 (small constant, see [1]). Try a larger K2 constant (e.g. 0.4) if you get a negative or NaN results. Returns: Full Value of Structural Similarity (SSIM) index. rr?AKernel size can't be greater than actual input size. Input size: . Kernel size: r(rrweightstridepaddinggroupsr')rr?r )rrrDconv2dr)rr5rAr9r:c1c2 n_channelsmu_xmu_ymu_xxmu_yymu_xysigma_xxsigma_yysigma_xyrOssrNrrrrI s,(    rIc#Csf|d}|d|dks|d|dkr'td|d||d}|d}|d}|d } |d} |d } tj||dd |d } tj| |dd |d } tj| |dd |d }tj| |dd |d }| d| d}|d|d}| || |}| || |}d }|d| d}| d| d}|| | | }|| | | }tj||dd |d |}tj||dd |d |}tj||dd |d |}tj||dd |d |}tj||fdd }tj||fdd }|d|||d|d||}|d|||d|d||} | |} | jdd }!|jdd }"|!|"fS)aCalculate Structural Similarity (SSIM) index for Complex X and Y per channel. Args: x: An input tensor. Shape :math:`(N, C, H, W, 2)`. y: A target tensor. Shape :math:`(N, C, H, W, 2)`. kernel: 2-D gauss kernel. k1: Algorithm parameter, K1 (small constant, see [1]). k2: Algorithm parameter, K2 (small constant, see [1]). Try a larger K2 constant (e.g. 0.4) if you get a negative or NaN results. Returns: Full Value of Complex Structural Similarity (SSIM) index. rr?rrdrer().r).rrrfr2r )r?rx) rrrDrkpowrstackr+r)#rr5rAr9r:rnrlrmx_realx_imagy_realy_imagmu1_realmu1_imagmu2_realmu2_imagmu1_sqmu2_sq mu1_mu2_real mu1_mu2_imag compensationx_sqy_sqx_y_realx_y_imag sigma1_sq sigma2_sq sigma12_real sigma12_imagsigma12mu1_mu2rMrLrNrOrrrrH=sL ( ,,  rH)r)rrN)r0r1r2rFTr3r4)r3r4)typingrrrrrtorch.nn.functionalnn functionalrDtorch.nn.modules.lossrrbrarr_rBr#r/r`rPrQrIrHrrrrs    +    Ds 7