U h;@spddlmZddlZddlmZddlmZddlZddlZ ddl m Z m Z m Z mZmZmZmZddlmZddlmZddlmmmZddlmZdd lmZmZd d d gZ Gd ddZ!Gddde!Z"Gddde!Z#GdddeZ$GdddZ%e edddddddd Z&dddddddZ'e edddddd d Z(e eddddd!d"d Z)dS)#) annotationsN)deepcopy)Literal) add_weightedclipclipped from_floatget_num_channelspreserve_channel_dimto_float)match_histograms)Protocol)PCA)MONO_CHANNEL_DIMENSIONSNUM_MULTI_CHANNEL_DIMENSIONSfourier_domain_adaptationapply_histogramadapt_pixel_distributionc@sZeZdZddddZdddddZdddd d Zdddd d Zdddd dZdS) BaseScalerNonereturncCs"d|_d|_d|_d|_d|_dSN)data_mindata_maxmeanvarscaleselfr ]/tmp/pip-unpacked-wheel-e8onvpoz/albumentations/augmentations/domain_adaptation/functional.py__init__s zBaseScaler.__init__ np.ndarrayxrcCstdSrNotImplementedErrorrr%r r r!fit szBaseScaler.fitcCstdSrr&r(r r r! transform#szBaseScaler.transformcCs||||Sr)r)r*r(r r r! fit_transform&s zBaseScaler.fit_transformcCstdSrr&r(r r r!inverse_transform*szBaseScaler.inverse_transformN)__name__ __module__ __qualname__r"r)r*r+r,r r r r!rs rcsVeZdZddddfdd Zdddd d Zdddd d Zdddd dZZS) MinMaxScalergg?ztuple[float, float]r) feature_rangercs(t|d|_|d|_d|_dS)Nr)superr"minmax data_range)rr2 __class__r r!r"/s   zMinMaxScaler.__init__r#r$cCsBtj|dd|_tj|dd|_|j|j|_d|j|jdk<dSNrZaxisr3)npr5rr6rr7r(r r r!r)5szMinMaxScaler.fitcCsL|jdks|jdks|jdkr&td||j|j}||j|j|jSNzpThis MinMaxScaler instance is not fitted yet. Call 'fit' with appropriate arguments before using this estimator.)rrr7 ValueErrorr6r5rr%Zx_stdr r r!r*<s zMinMaxScaler.transformcCsL|jdks|jdks|jdkr&td||j|j|j}||j|jSr=)rrr7r>r5r6r?r r r!r,Es zMinMaxScaler.inverse_transform)r1r-r.r/r"r)r*r, __classcell__r r r8r!r0.s r0csReZdZddfdd ZdddddZdddd d Zdddd d ZZS) StandardScalerrrcstdSr)r4r"rr8r r!r"PszStandardScaler.__init__r#r$cCsBtj|dd|_tj|dd|_t|j|_d|j|jdk<dSr:)r<rrsqrtrr(r r r!r)SszStandardScaler.fitcCs,|jdks|jdkrtd||j|jSNzrThis StandardScaler instance is not fitted yet. Call 'fit' with appropriate arguments before using this estimator.rrr>r(r r r!r*Zs zStandardScaler.transformcCs,|jdks|jdkrtd||j|jSrDrEr(r r r!r,bs z StandardScaler.inverse_transformr@r r r8r!rBOsrBc@sVeZdZejdddddZejd dddddd Zejd ddddd d ZdS)TransformerInterfacer#r$cCsdSrr r(r r r!r,lsz&TransformerInterface.inverse_transformNznp.ndarray | None)r%yrcCsdSrr rr%rGr r r!r)oszTransformerInterface.fitcCsdSrr rHr r r!r*rszTransformerInterface.transform)N)N)r-r.r/abcabstractmethodr,r)r*r r r r!rFks rFc@seZdZdddddddZdddd d Zdddd d Zdddd dZdddddddZedddddZ dddddZ dS) DomainAdapterNNrFr#ztuple[(None, None)]) transformerref_imgcolor_conversionscCs<|\|_|_t||_||_t||_|j||dSr) color_in color_outrsource_transformertarget_transformerr num_channelsr)flatten)rrMrNrOr r r!r"ws    zDomainAdapter.__init__)imgrcCs|jdkr|St||jSr)rPcv2cvtColorrrVr r r! to_colorspaceszDomainAdapter.to_colorspacecCs$|jdkr|Stt|tj|jSr)rQrWrXrr<uint8rYr r r!from_colorspaces zDomainAdapter.from_colorspacecCs ||}t|}|d|jS)N)rZr reshaperTrYr r r!rUs zDomainAdapter.flattenint)pixelsheightwidthrcCsJt|dddd}|jdkr4||||S|||||jS)Nrr3r[)r<rastyperTr\r^)rr`rarbr r r! reconstructs zDomainAdapter.reconstructr$cCstt|jSr)r<signtrace components_)r%r r r! _pca_signszDomainAdapter._pca_sign)imagercCs|jdd\}}||}|j|t|jdrht|jdrh||j||jkrh|jjd9_|j|}|j |}| |||S)Nrhr]) shaperUrRr)hasattrrSrirhr*r,re)rrjrarbr`Zrepresentationresultr r r!__call__s      zDomainAdapter.__call__N)rL) r-r.r/r"rZr\rUre staticmethodriror r r r!rKvs rKr#z&Literal[('pca', 'standard', 'minmax')]float)rVreftransform_typeweightrc Cs|j|jkrtdt|}t|}||kr4td|dkrPt|}t|}|j|jkrxtj||jddtjd}|j}|tj krt |tj }t |tj }t t td|}t||d}||tj } |tj d|| |} |tj kr| St| S)Nz9Input image and reference image must have the same dtype.zFInput image and reference image must have the same number of channels.r3rk)dsize interpolation)ZpcastandardZminmax)rMrN)Zdtyper>r r<squeezerlrWresizeZ INTER_AREAfloat32rr[rrBr0rKrdr ) rVrrrsrtZimg_num_channelsZref_num_channelsZoriginal_dtyperMadapterZ transformedrnr r r!rs(        )amp_srcamp_trgbetarc Cs|jdd}ttt||}t|\}}|\}}tdt||tt|||} } tdt||tt|||} } || | | | f|| | | | f<|S)Nrkr)rlr_r<floorr5 fgeometriccenterr6) r|r}r~Z image_shapeZborderZcenter_xZcenter_yrarbh1h2Zw1Zw2r r r!low_freq_mutates&&$r)rV target_imgr~rcCs<|tj}|tj}t|jtkr4tj|dd}t|jtkrPtj|dd}|jd}t|}t|D]}tj |dddd|f}tj |dddd|f} tj |} tj | } t | t | } } t | }t| ||}tj |td| }tj |}t||dddd|f<ql|S)a Apply Fourier Domain Adaptation to the input image using a target image. This function performs domain adaptation in the frequency domain by modifying the amplitude spectrum of the source image based on the target image's amplitude spectrum. It preserves the phase information of the source image, which helps maintain its content while adapting its style to match the target image. Args: img (np.ndarray): The source image to be adapted. Can be grayscale or RGB. target_img (np.ndarray): The target image used as a reference for adaptation. Should have the same dimensions as the source image. beta (float): The adaptation strength, typically in the range [0, 1]. Higher values result in stronger adaptation towards the target image's style. Returns: np.ndarray: The adapted image with the same shape and type as the input image. Raises: ValueError: If the source and target images have different shapes. Note: - Both input images are converted to float32 for processing. - The function handles both grayscale (2D) and color (3D) images. - For grayscale images, an extra dimension is added to facilitate uniform processing. - The adaptation is performed channel-wise for color images. - The output is clipped to the valid range and preserves the original number of channels. The adaptation process involves the following steps for each channel: 1. Compute the 2D Fourier Transform of both source and target images. 2. Shift the zero frequency component to the center of the spectrum. 3. Extract amplitude and phase information from the source image's spectrum. 4. Mutate the source amplitude using the target amplitude and the beta parameter. 5. Combine the mutated amplitude with the original phase. 6. Perform the inverse Fourier Transform to obtain the adapted channel. The `low_freq_mutate` function (not shown here) is responsible for the actual amplitude mutation, focusing on low-frequency components which carry style information. Example: >>> import numpy as np >>> import albumentations as A >>> source_img = np.random.rand(100, 100, 3).astype(np.float32) >>> target_img = np.random.rand(100, 100, 3).astype(np.float32) >>> adapted_img = A.fourier_domain_adaptation(source_img, target_img, beta=0.5) >>> assert adapted_img.shape == source_img.shape References: - "FDA: Fourier Domain Adaptation for Semantic Segmentation" (Yang and Soatto, 2020, CVPR) https://openaccess.thecvf.com/content_CVPR_2020/papers/Yang_FDA_Fourier_Domain_Adaptation_for_Semantic_Segmentation_CVPR_2020_paper.pdf r]r;Ny?)rdr<rzlenrlrZ expand_dimsZ zeros_likerangeZfftZfft2ZfftshiftabsZanglercopyZ ifftshiftexpZifft2real)rVrr~Zsrc_imgZtrg_imgrTZ src_in_trgZ channel_idZfft_srcZfft_trgZfft_src_shiftedZfft_trg_shiftedr|Zpha_srcr}Z mutated_ampZfft_src_mutatedZsrc_in_trg_channelr r r!rs(6         )rVreference_image blend_ratiorcCs|jdd|jddkr:tj||jd|jdfd}t|}t|}t|||jtkrp|jddkrpdndd}t|||d|S)aApply histogram matching to an input image using a reference image and blend the result. This function performs histogram matching between the input image and a reference image, then blends the result with the original input image based on the specified blend ratio. Args: img (np.ndarray): The input image to be transformed. Can be either grayscale or RGB. Supported dtypes: uint8, float32 (values should be in [0, 1] range). reference_image (np.ndarray): The reference image used for histogram matching. Should have the same number of channels as the input image. Supported dtypes: uint8, float32 (values should be in [0, 1] range). blend_ratio (float): The ratio for blending the matched image with the original image. Should be in the range [0, 1], where 0 means no change and 1 means full histogram matching. Returns: np.ndarray: The transformed image after histogram matching and blending. The output will have the same shape and dtype as the input image. Supported image types: - Grayscale images: 2D arrays - RGB images: 3D arrays with 3 channels - Multispectral images: 3D arrays with more than 3 channels Note: - If the input and reference images have different sizes, the reference image will be resized to match the input image's dimensions. - The function uses `match_histograms` from scikit-image for the core histogram matching. - The @clipped and @preserve_channel_dim decorators ensure the output is within the valid range and maintains the original number of dimensions. Example: >>> import numpy as np >>> from albumentations.augmentations.domain_adaptation_functional import apply_histogram >>> input_image = np.random.randint(0, 256, (100, 100, 3), dtype=np.uint8) >>> reference_image = np.random.randint(0, 256, (100, 100, 3), dtype=np.uint8) >>> result = apply_histogram(input_image, reference_image, blend_ratio=0.7) Nrkr3r)ru)Z channel_axis) rlrWryr<rxr ndimrr)rVrrmatchedr r r!r?s)  )* __future__rrIrrtypingrrWZnumpyr<Zalbucorerrrrr r r Zskimage.exposurer Ztyping_extensionsr Z1albumentations.augmentations.geometric.functionalZ augmentationsZ geometricZ functionalrZ"albumentations.augmentations.utilsrZalbumentations.core.typesrr__all__rr0rBrFrKrrrrr r r r!s<   $   ! 5$]