o oiI@sddlmZddlZddlmZddlmZmZddlmZdej dee dej fd d Z ddej dej de de de de dej fddZGdddejZdS))ListN)nn)filter2d_separableget_gaussian_kernel1d)_compute_paddingimgcropping_shapereturncCs0tjj||d |d |d |d fS)z.Crop out the part of "valid" convolution area.r)torchr functionalpad)rrrG/home/ubuntu/.local/lib/python3.10/site-packages/kornia/metrics/ssim.py_crops$r?-q=sameimg1img2 window_sizemax_valepspaddingcCst|tjstdt|t|tjstdt|t|ts,tdt|t|jdks;td|jt|jdksJtd|j|j|jks\td|jd|jt |d |j |j d }d |d }d |d }t |||} t |||} g} |dkr|jd} } t | | g} t| | } t| | } n|dkr | d }| d }| | }t |d ||}t |d ||}t ||||}|dkrt|| }t|| }t|| }n|dkr ||}||}||}d||d||}||||||}|||S)aOCompute the Structural Similarity (SSIM) index map between two images. Measures the (SSIM) index between each element in the input `x` and target `y`. The index can be described as: .. math:: \text{SSIM}(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:`c_1=(k_1 L)^2` and :math:`c_2=(k_2 L)^2` are two variables to stabilize the division with weak denominator. - :math:`L` is the dynamic range of the pixel-values (typically this is :math:`2^{\#\text{bits per pixel}}-1`). Args: img1: the first input image with shape :math:`(B, C, H, W)`. img2: the second input image with shape :math:`(B, C, H, W)`. window_size: the size of the gaussian kernel to smooth the images. max_val: the dynamic range of the images. eps: Small value for numerically stability when dividing. padding: ``'same'`` | ``'valid'``. Whether to only use the "valid" convolution area to compute SSIM to match the MATLAB implementation of original SSIM paper. Returns: The ssim index map with shape :math:`(B, C, H, W)`. Examples: >>> input1 = torch.rand(1, 4, 5, 5) >>> input2 = torch.rand(1, 4, 5, 5) >>> ssim_map = ssim(input1, input2, 5) # 1x4x5x5 z+Input img1 type is not a torch.Tensor. Got z+Input img2 type is not a torch.Tensor. Got z'Input max_val type is not a float. Got z,Invalid img1 shape, we expect BxCxHxW. Got: z,Invalid img2 shape, we expect BxCxHxW. Got: z,img1 and img2 shapes must be the same. Got: z and g?)devicedtypeg{Gz?r gQ?validrg@) isinstancer Tensor TypeErrortypefloatlenshape ValueErrorrrrrrr)rrrrrrkernelC1C2mu1mu2rheightwidthmu1_sqmu2_sqmu1_mu2 mu_img1_sq mu_img2_sq mu_img1_img2 sigma1_sq sigma2_sqsigma12numdenrrrssim"sV +              r;c sTeZdZdZddedededed d f fd d Zd ej dej d ej fddZ Z S)SSIMa'Create a module that computes the Structural Similarity (SSIM) index between two images. Measures the (SSIM) index between each element in the input `x` and target `y`. The index can be described as: .. math:: \text{SSIM}(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:`c_1=(k_1 L)^2` and :math:`c_2=(k_2 L)^2` are two variables to stabilize the division with weak denominator. - :math:`L` is the dynamic range of the pixel-values (typically this is :math:`2^{\#\text{bits per pixel}}-1`). Args: window_size: the size of the gaussian kernel to smooth the images. max_val: the dynamic range of the images. eps: Small value for numerically stability when dividing. padding: ``'same'`` | ``'valid'``. Whether to only use the "valid" convolution area to compute SSIM to match the MATLAB implementation of original SSIM paper. Shape: - Input: :math:`(B, C, H, W)`. - Target :math:`(B, C, H, W)`. - Output: :math:`(B, C, H, W)`. Examples: >>> input1 = torch.rand(1, 4, 5, 5) >>> input2 = torch.rand(1, 4, 5, 5) >>> ssim = SSIM(5) >>> ssim_map = ssim(input1, input2) # 1x4x5x5 rrrrrrrr Ncs&t||_||_||_||_dSN)super__init__rrrr)selfrrrr __class__rrr?s  z SSIM.__init__rrcCst|||j|j|j|jSr=)r;rrrr)r@rrrrrforwardsz SSIM.forwardrrr) __name__ __module__ __qualname____doc__intr%strr?r r"rC __classcell__rrrArr<s$%$r<rD)typingrr rkornia.filtersrrkornia.filters.filterrr"rIrr%rJr;Moduler<rrrrs2     l