o oi%@sddlmZmZmZmZmZmZddlmZ ddl m Z ddl m Z ddlmZmZddlmZddlmZmZddlmZGd d d e ZGd d d e Zd S))AnyDictListOptionalTupleUnion)random_generator)GeometricAugmentationBase2D)Resample)Tensor as_tensor)affine)_compute_rotation_matrix_compute_tensor_center)eye_likecs eZdZdZejjddddfdeee e e e fe e fdee e efdeded e d ed d ffd d Zdedee efdee efd efddZ ddedee efdee efdeed ef ddZ ddedee efdeedee e e fd ef ddZZS)RandomRotationaApply a random rotation to a tensor image or a batch of tensor images given an amount of degrees. .. image:: _static/img/RandomRotation.png Args: degrees: range of degrees to select from. If degrees is a number the range of degrees to select from will be (-degrees, +degrees). resample: Default: the interpolation mode. same_on_batch: apply the same transformation across the batch. align_corners: interpolation flag. p: probability of applying the transformation. keepdim: whether to keep the output shape the same as input (True) or broadcast it to the batch form (False). Shape: - Input: :math:`(C, H, W)` or :math:`(B, C, H, W)`, Optional: :math:`(B, 3, 3)` - Output: :math:`(B, C, H, W)` .. note:: This function internally uses :func:`kornia.geometry.transform.affine`. Examples: >>> rng = torch.manual_seed(0) >>> input = torch.tensor([[1., 0., 0., 2.], ... [0., 0., 0., 0.], ... [0., 1., 2., 0.], ... [0., 0., 1., 2.]]) >>> aug = RandomRotation(degrees=45.0, p=1.) >>> out = aug(input) >>> out tensor([[[[0.9824, 0.0088, 0.0000, 1.9649], [0.0000, 0.0029, 0.0000, 0.0176], [0.0029, 1.0000, 1.9883, 0.0000], [0.0000, 0.0088, 1.0117, 1.9649]]]]) >>> aug.transform_matrix tensor([[[ 1.0000, -0.0059, 0.0088], [ 0.0059, 1.0000, -0.0088], [ 0.0000, 0.0000, 1.0000]]]) >>> inv = aug.inverse(out) To apply the exact augmenation again, you may take the advantage of the previous parameter state: >>> input = torch.randn(1, 3, 32, 32) >>> aug = RandomRotation(degrees=45.0, p=1.) >>> (aug(input) == aug(input, params=aug._params)).all() tensor(True) FT?degreesresample same_on_batch align_cornerspkeepdimreturnNc<tj|||dt|dddf|_t||d|_dS)Nrrrr)gvgv@rrsuper__init__rgPlainUniformGenerator_param_generatorr getflags)selfrrrrrr __class__^/home/ubuntu/.local/lib/python3.10/site-packages/kornia/augmentation/_2d/geometric/rotation.pyr P zRandomRotation.__init__inputparamsr%cCsx|d|}t|}t|||jdd}td|dd}|dddf|dddf<|dddf|dddf<|S)NrrF shared_memory)torrexpandshaperr&r,r-r%anglescenter rotation_mat trans_matr)r)r*compute_transformation^sz%RandomRotation.compute_transformation transformcCNt|tstdt|dt||dddddf|djd|dS N*Expected the `transform` be a Tensor. Got ..r/rzerosr isinstancer TypeErrortyper namelowerr&r,r-r%r<r)r)r*apply_transforml 0zRandomRotation.apply_transformsizecC@t|tstdt|d|j||jt||j|jd|dSNr?r@)devicedtype)r-r<r% rDr rErFrJ_paramsr rOrPr&r,r%r<rLr)r)r*inverse_transformt z RandomRotation.inverse_transformNNN)__name__ __module__ __qualname____doc__r BILINEARrGrr floatrrstrintboolr rrr;rrJrT __classcell__r)r)r'r*rs`5 *     rcseZdZdZejjddddfdeeefde e eefde de d e d e d d ffd d Z dedee efdee efd efddZ ddedee efdee efdeed ef ddZ ddedee efdeedeeeefd ef ddZZS)RandomRotation90aTApply a random 90 * n degree rotation to a tensor image or a batch of tensor images. Args: times: the range of n times 90 degree rotation needs to be applied. resample: Default: the interpolation mode. same_on_batch: apply the same transformation across the batch. align_corners: interpolation flag. p: probability of applying the transformation. keepdim: whether to keep the output shape the same as input (True) or broadcast it to the batch form (False). Shape: - Input: :math:`(C, H, W)` or :math:`(B, C, H, W)`, Optional: :math:`(B, 3, 3)` - Output: :math:`(B, C, H, W)` .. note:: This function internally uses :func:`kornia.geometry.transform.affine`. This version is relatively slow as it operates based on affine transformations. Examples: >>> rng = torch.manual_seed(1) >>> torch.set_printoptions(sci_mode=False) >>> input = torch.tensor([[1., 0., 0., 2.], ... [0., 0., 0., 0.], ... [0., 1., 2., 0.], ... [0., 0., 1., 2.]]) >>> aug = RandomRotation90(times=(1, 1), p=1.) >>> out = aug(input) >>> out tensor([[[[ 2.0000, 0.0000, 0.0000, 2.0000], [ 0.0000, 0.0000, 2.0000, 1.0000], [ 0.0000, 0.0000, 1.0000, 0.0000], [ 1.0000, 0.0000, 0.0000, 0.0000]]]]) >>> aug.transform_matrix tensor([[[ -0.0000, 1.0000, 0.0000], [ -1.0000, -0.0000, 3.0000], [ 0.0000, 0.0000, 1.0000]]]) >>> inv = aug.inverse(out) >>> torch.set_printoptions(profile='default') To apply the exact augmenation again, you may take the advantage of the previous parameter state: >>> input = torch.randn(1, 3, 32, 32) >>> aug = RandomRotation90(times=(-1, 1), p=1.) >>> (aug(input) == aug(input, params=aug._params)).all() tensor(True) FTrtimesrrrrrrNcr)Nrrcr)r/rr)r&rcrrrrrr'r)r*r r+zRandomRotation90.__init__r,r-r%cCsd|d|}t|}t|||jdd}td|dd}|dddf|dddf<|dddf|dddf<|S) NgV@rcrr.r/Fr0r2)roundr3rrr4r5rr6r)r)r*r;sz'RandomRotation90.compute_transformationr<cCr=r>rCrIr)r)r*rJrKz RandomRotation90.apply_transformrLcCrMrNrQrSr)r)r*rTrUz"RandomRotation90.inverse_transformrVrW)rXrYrZr[r r\rGtupler_rr^r`r]r r dictrr;rrJrTrar)r)r'r*rbs`3  *     rbN)typingrrrrrrkornia.augmentationrr!&kornia.augmentation._2d.geometric.baser kornia.constantsr kornia.corer r kornia.geometry.transformr !kornia.geometry.transform.affwarprrkornia.utils.miscrrrbr)r)r)r*s      h