a PSicSn@sddlZddlmZddlmZmZmZmZddlZddlm Z ddl m Z m Z gdZe eee eeedd d ZGd d d eZGd ddejjZGdddejjZGdddejjZGdddejjZdS)N)Enum)ListTupleOptionalDict)Tensor) functionalInterpolationMode)AutoAugmentPolicy AutoAugment RandAugmentTrivialAugmentWideAugMiximgop_name magnitude interpolationfillc Cs|dkr>tj|dddgdtt|dg||ddgd}n|dkr|tj|dddgddtt|g||ddgd}n|dkrtj|dt|dgd|ddg|d}nP|d krtj|ddt|gd|ddg|d}n |d krtj||||d }n|d krt|d|}n|d kr2t|d|}n|dkrNt |d|}n|dkrjt |d|}n|dkrt |t|}nv|dkrt ||}n^|dkrt |}nH|dkrt|}n2|dkrt|}n|dkrntd|d|S)NShearXr?)angle translatescaleshearrrcenterShearY TranslateX)rrrrrr TranslateYRotaterr BrightnessColorContrast Sharpness PosterizeSolarize AutoContrastEqualizeInvertIdentityzThe provided operator  is not recognized.)Faffinemathdegreesatanintrotateadjust_brightnessadjust_saturationadjust_contrastadjust_sharpness posterizesolarize autocontrastequalizeinvert ValueErrorrr?^/var/www/html/django/DPS/env/lib/python3.9/site-packages/torchvision/transforms/autoaugment.py _apply_op s                   rAc@seZdZdZdZdZdZdS)r zoAutoAugment policies learned on different datasets. Available policies are IMAGENET, CIFAR10 and SVHN. imagenetcifar10svhnN)__name__ __module__ __qualname____doc__IMAGENETCIFAR10SVHNr?r?r?r@r ]sr c seZdZdZejejdfeeee e ddfdd Z ee e e e e eefe e e eeffdddZee eefee e eeffd d d Zeee eeefd d dZeedddZe dddZZS)r a?AutoAugment data augmentation method based on `"AutoAugment: Learning Augmentation Strategies from Data" `_. If the image is torch Tensor, it should be of type torch.uint8, and it is expected to have [..., 1 or 3, H, W] shape, where ... means an arbitrary number of leading dimensions. If img is PIL Image, it is expected to be in mode "L" or "RGB". Args: policy (AutoAugmentPolicy): Desired policy enum defined by :class:`torchvision.transforms.autoaugment.AutoAugmentPolicy`. Default is ``AutoAugmentPolicy.IMAGENET``. interpolation (InterpolationMode): Desired interpolation enum defined by :class:`torchvision.transforms.InterpolationMode`. Default is ``InterpolationMode.NEAREST``. If input is Tensor, only ``InterpolationMode.NEAREST``, ``InterpolationMode.BILINEAR`` are supported. fill (sequence or number, optional): Pixel fill value for the area outside the transformed image. If given a number, the value is used for all bands respectively. N)policyrrreturncs,t||_||_||_|||_dSN)super__init__rLrr _get_policiespolicies)selfrLrr __class__r?r@rPys  zAutoAugment.__init__)rLrMcCsJ|tjkrgdS|tjkr$gdS|tjkr6gdStd|ddS)N)))r'皙?)r!333333? )r(rXr)rXNr*皙?Nr*rXN))r'rX)r'rXr*rVN)r(皙?)rdr!r_rW))r(rXr`))r'r_r[r*rN))r!rerh)r(rXrW)r`)r'rVrb)rgr$rVr))r!rVrYr`))r*rNr^r+rXNri)r$rXrf)r%rrW)rg)r$r))r$r_rW)r(r_ra))r&rVrarl))rrXr[ri)rjr`rcrZrkrmr]))r+皙?N)r%rerb))r!ffffff?rn)r333333?rY))r&r_r)r&?rh))r?rWr rqrY))r)rtNr*rsN))rrera)r'rrra))r$rVrh)r#rXra))r&rrrY)r#rqrY)r`)r*rtN))r%rXra)r&rXr[))r$rqra)rrtrW))r*rrN)r)rVN))r rVrh)r&rerb))r#rsrb)r$rerW))r(rtrn)r+rN)r*reNr\)rwr`))r$rsrYr`)r)r_N)r(rerW))r#rprh)r$rqr))r(rVr[r)rsN))r rsrYru)ry)r(r_rh)r^ro)rury))rrsrf)r+reN)rrsrWr+rqN)r`)r(rXrbr+rsNr`r`)r!rsrh)rzrx)r{)r+rVN))rrsr[)r(rerb)r~rxr)rz)r(rrrh))rr_rWr|)rv)r rXrbr}))r%rrrhr!r_rf)r+r_N)r rrn))rrqrb)r(rVrW)rlr))rrrra)rrsrh))rrprbrl))r(rqrn)r rXra))rr_rfr))rrqrY)r r_rh))rr_r[)r)rqN))rrqrnrozThe provided policy r-)r rIrJrKr>)rSrLr?r?r@rQs   zAutoAugment._get_policiesnum_bins image_sizerMcCstdd|dftdd|dftdd|d|dftdd|d|dftdd|dftdd|dftdd|dftdd|dftdd|dfd t||dd d ftd d|d ftdd ftdd ftdd fd S)NrrrTt?rr>@rsrWrfFo@)rrrr r!r#r$r%r&r'r(r)r*r+)torchlinspacearangeroundr3tensorrSrrr?r?r@_augmentation_spaces$   zAutoAugment._augmentation_space) transform_numrMcCs4tt|d}td}tdd}|||fS)zGet parameters for autoaugment transformation Returns: params required by the autoaugment transformation r)rnrn)r3rrandintitemrand)rZ policy_idprobssignsr?r?r@ get_paramss  zAutoAugment.get_paramsrrMcCs|j}t|\}}}t|trTt|ttfr>t|g|}n|durTdd|D}|t|j \}}}| d||f} t |j |D]n\} \} } } || | kr| | \}}| durt||  nd}|r|| dkr|d9}t || ||j|d}q|S) z img (PIL Image or Tensor): Image to be transformed. Returns: PIL Image or Tensor: AutoAugmented image. NcSsg|] }t|qSr?float.0fr?r?r@ z'AutoAugment.forward.. rrr")rr.get_dimensions isinstancerr3rrlenrRr enumeraterrAr)rSrrchannelsheightwidthZ transform_idrrop_metairpZ magnitude_id magnitudessignedrr?r?r@forwards"   zAutoAugment.forwardrMcCs|jjd|jd|jdS)Nz(policy=, fill=))rUrErLr)rSr?r?r@__repr__szAutoAugment.__repr__)rErFrGrHr rIr NEARESTrrrrPrstrr3rQrrboolr staticmethodrrr __classcell__r?r?rTr@r hs$  * Z* r cseZdZdZdddejdfeeeeeee ddfdd Z ee eefe e e eeffd d d Zeed d dZe dddZZS)r a~RandAugment data augmentation method based on `"RandAugment: Practical automated data augmentation with a reduced search space" `_. If the image is torch Tensor, it should be of type torch.uint8, and it is expected to have [..., 1 or 3, H, W] shape, where ... means an arbitrary number of leading dimensions. If img is PIL Image, it is expected to be in mode "L" or "RGB". Args: num_ops (int): Number of augmentation transformations to apply sequentially. magnitude (int): Magnitude for all the transformations. num_magnitude_bins (int): The number of different magnitude values. interpolation (InterpolationMode): Desired interpolation enum defined by :class:`torchvision.transforms.InterpolationMode`. Default is ``InterpolationMode.NEAREST``. If input is Tensor, only ``InterpolationMode.NEAREST``, ``InterpolationMode.BILINEAR`` are supported. fill (sequence or number, optional): Pixel fill value for the area outside the transformed image. If given a number, the value is used for all bands respectively. rnrYN)num_opsrnum_magnitude_binsrrrMcs,t||_||_||_||_||_dSrN)rOrPrrrrr)rSrrrrrrTr?r@rP2s  zRandAugment.__init__rcCstddftdd|dftdd|dftdd|d|dftdd|d|dftdd|dftdd |dftdd |dftdd |dftdd |dfd t||dd dftd d|dftddftddfd S)NrFrrTrrrrrsrWrfrr,rrrr r!r#r$r%r&r'r(r)r*rrrrrr3rr?r?r@rAs $  zRandAugment._augmentation_spacerc Cs|j}t|\}}}t|trTt|ttfr>t|g|}n|durTdd|D}||j||f}t |j D]}tt t |d}t||} || \} } | jdkrt| |jnd} | rt ddr| d9} t|| | |j|d }qp|S) img (PIL Image or Tensor): Image to be transformed. Returns: PIL Image or Tensor: Transformed image. NcSsg|] }t|qSr?rrr?r?r@rarz'RandAugment.forward..rrrrnrr")rr.rrrr3rrrrangerrrrrlistkeysndimrrAr) rSrrrrrr_op_indexrrrrr?r?r@rTs"   zRandAugment.forwardrc Cs:|jjd|jd|jd|jd|jd|jd }|S)Nz (num_ops=z , magnitude=z, num_magnitude_bins=, interpolation=rr)rUrErrrrrrSsr?r?r@ros  zRandAugment.__repr__)rErFrGrHr rr3rrrrPrrrrrrrrrr?r?rTr@r s" *r cs|eZdZdZdejdfeeeee ddfdd Z ee e e eeffddd Zeed d d Ze d ddZZS)raDataset-independent data-augmentation with TrivialAugment Wide, as described in `"TrivialAugment: Tuning-free Yet State-of-the-Art Data Augmentation" `_. If the image is torch Tensor, it should be of type torch.uint8, and it is expected to have [..., 1 or 3, H, W] shape, where ... means an arbitrary number of leading dimensions. If img is PIL Image, it is expected to be in mode "L" or "RGB". Args: num_magnitude_bins (int): The number of different magnitude values. interpolation (InterpolationMode): Desired interpolation enum defined by :class:`torchvision.transforms.InterpolationMode`. Default is ``InterpolationMode.NEAREST``. If input is Tensor, only ``InterpolationMode.NEAREST``, ``InterpolationMode.BILINEAR`` are supported. fill (sequence or number, optional): Pixel fill value for the area outside the transformed image. If given a number, the value is used for all bands respectively. rN)rrrrMcs t||_||_||_dSrN)rOrPrrr)rSrrrrTr?r@rPs zTrivialAugmentWide.__init__)rrMcCstddftdd|dftdd|dftdd|dftdd|dftdd|dftdd|dftdd|dftdd|dftdd|dfdt||dd dftd d|dftddftddfd S) NrFgGz?Tg@@g`@rWrrbrrr)rSrr?r?r@rs $  z&TrivialAugmentWide._augmentation_spacerc Cs|j}t|\}}}t|trTt|ttfr>t|g|}n|durTdd|D}||j}tt t |d }t ||}||\} } | jdkrt| t j t | dt jd nd} | rt ddr| d 9} t||| |j|d S) rNcSsg|] }t|qSr?rrr?r?r@rrz.TrivialAugmentWide.forward..rrdtyperrnrr")rr.rrrr3rrrrrrrrrrlongrAr) rSrrrrrrrrrrrr?r?r@rs$   $zTrivialAugmentWide.forwardrcCs*|jjd|jd|jd|jd}|S)Nz(num_magnitude_bins=rrr)rUrErrrrr?r?r@rs zTrivialAugmentWide.__repr__)rErFrGrHr rr3rrrrPrrrrrrrrrr?r?rTr@r|s  rc seZdZdZdddddejdfeeeeeee e eddfdd Z ee eefe ee eeffd d d Zejjed ddZejjedddZeedddZeedddZed ddZZS)raAugMix data augmentation method based on `"AugMix: A Simple Data Processing Method to Improve Robustness and Uncertainty" `_. If the image is torch Tensor, it should be of type torch.uint8, and it is expected to have [..., 1 or 3, H, W] shape, where ... means an arbitrary number of leading dimensions. If img is PIL Image, it is expected to be in mode "L" or "RGB". Args: severity (int): The severity of base augmentation operators. Default is ``3``. mixture_width (int): The number of augmentation chains. Default is ``3``. chain_depth (int): The depth of augmentation chains. A negative value denotes stochastic depth sampled from the interval [1, 3]. Default is ``-1``. alpha (float): The hyperparameter for the probability distributions. Default is ``1.0``. all_ops (bool): Use all operations (including brightness, contrast, color and sharpness). Default is ``True``. interpolation (InterpolationMode): Desired interpolation enum defined by :class:`torchvision.transforms.InterpolationMode`. Default is ``InterpolationMode.NEAREST``. If input is Tensor, only ``InterpolationMode.NEAREST``, ``InterpolationMode.BILINEAR`` are supported. fill (sequence or number, optional): Pixel fill value for the area outside the transformed image. If given a number, the value is used for all bands respectively. rhrTN)severity mixture_width chain_depthalphaall_opsrrrMcsntd|_d|kr&|jks@ntd|jd|d||_||_||_||_||_||_ ||_ dS)Nrrz!The severity must be between [1, z]. Got z instead.) rOrP_PARAMETER_MAXr>rrrrrrr)rSrrrrrrrrTr?r@rPs zAugMix.__init__rc Cs tdd|dftdd|dftd|dd|dftd|dd|dftdd|dfdt||ddd ftd d|d ftdd ftdd fd }|jr|tdd |dftdd |dftdd |dftdd |dfd |S)NrrrTrg@rrrfFr) rrrr r!r'r(r)r*rs)r#r$r%r&)rrrrr3rrupdate)rSrrrr?r?r@rs&$   zAugMix._augmentation_spacercCs t|SrN)r. pil_to_tensorrSrr?r?r@_pil_to_tensorszAugMix._pil_to_tensor)rcCs t|SrN)r. to_pil_imagerr?r?r@_tensor_to_pilszAugMix._tensor_to_pil)paramsrMcCs t|SrN)r_sample_dirichlet)rSrr?r?r@rszAugMix._sample_dirichlet)orig_imgrMc Cs||j}t|\}}}t|trZ|}t|ttfrBt|g|}qd|durddd|D}n ||}||j ||f}t |j }| dgt d|jd|} | dgdg| jd} |tj|j|jg| jd| dd} |tj|jg|j| jd| dd| dddf | ddg} | dddf | | } t|jD]}| }|jdkrn|jnttjddd d }t|D]}ttt|d }t ||}||\}}|jdkrt|tj|jd tjd nd }|rtd d r|d9}t||||j |d}q| !| dd|f | |qT| |j"|j#d } t|tsx|$| S| S)rNcSsg|] }t|qSr?rrr?r?r@r/rz"AugMix.forward..rrfr)devicerr)lowhighsizerrrnrr")%rr.rrrr3rrrrrshapeviewmaxrrrrrrrexpandrrrrrrrrrrAradd_torr)rSrrrrrrrZ orig_dimsbatch batch_dimsmZcombined_weightsmixraugdepthrrrrrrr?r?r@r!sR    "$*  $$  zAugMix.forwardcCsJ|jjd|jd|jd|jd|jd|jd|jd|jd}|S) Nz (severity=z, mixture_width=z, chain_depth=z, alpha=z , all_ops=rrr) rUrErrrrrrrrr?r?r@r[s"  zAugMix.__repr__)rErFrGrHr BILINEARr3rrrrrPrrrrrrjitunusedrrrrrrr?r?rTr@rs4 *:r)r0enumrtypingrrrrrrr r.r __all__rrrAr nnModuler r rrr?r?r?r@s   P 8]V