a ¨þdSnã@sÚddlZddlmZddlmZmZmZmZddlZddlm Z ddl m Z m Z gd¢Ze eee eeedœd d „ZGd d „d eƒZGd d„dejjƒZGdd„dejjƒZGdd„dejjƒZGdd„dejjƒZdS)éN)ÚEnum)ÚDictÚListÚOptionalÚTuple)ÚTensoré)Ú functionalÚInterpolationMode)ÚAutoAugmentPolicyÚ AutoAugmentÚ RandAugmentÚTrivialAugmentWideÚAugMix©ÚimgÚop_nameÚ magnitudeÚ interpolationÚfillc Cs|dkr>tj|dddgdt t |¡¡dg||ddgd}n¾|dkr|tj|dddgddt t |¡¡g||ddgd}n€|dkr¬tj|dt|ƒdgd|ddg|d}nP|d krÜtj|ddt|ƒgd|ddg|d}n |d krútj||||d }n|d krt |d|¡}næ|d kr2t |d|¡}nÊ|dkrNt  |d|¡}n®|dkrjt  |d|¡}n’|dkr†t  |t|ƒ¡}nv|dkržt  ||¡}n^|dkr´t  |¡}nH|dkrÊt |¡}n2|dkràt |¡}n|dkrìntd|›dƒ‚|S)NÚShearXçrçð?)ÚangleÚ translateÚscaleÚshearrrÚcenterÚShearYÚ TranslateX)rrrrrrÚ TranslateYÚRotate©rrÚ BrightnessÚColorÚContrastÚ SharpnessÚ PosterizeÚSolarizeÚ AutoContrastÚEqualizeÚInvertÚIdentityzThe provided operator ú is not recognized.)ÚFZaffineÚmathÚdegreesÚatanÚintÚrotateZadjust_brightnessZadjust_saturationZadjust_contrastZadjust_sharpnessZ posterizeZsolarizeZ autocontrastZequalizeÚinvertÚ ValueErrorr©r6úk/var/www/html/stable-diffusion-webui/venv/lib/python3.9/site-packages/torchvision/transforms/autoaugment.pyÚ _apply_op s„ø ø  ù  ù              r8c@seZdZdZdZdZdZdS)r zoAutoAugment policies learned on different datasets. Available policies are IMAGENET, CIFAR10 and SVHN. ZimagenetZcifar10ZsvhnN)Ú__name__Ú __module__Ú __qualname__Ú__doc__ÚIMAGENETÚCIFAR10ÚSVHNr6r6r6r7r ]sr c sÞeZdZdZejejdfeeee e ddœ‡fdd„ Z ee e e e e eefe e e eeffdœdd„Zee eefee e eeffd œd d „Zeee eeefd œd d„ƒZeedœdd„Ze dœdd„Z‡ZS)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)ÚpolicyrrÚreturncs,tƒ ¡||_||_||_| |¡|_dS©N)ÚsuperÚ__init__r@rrÚ _get_policiesÚpolicies)Úselfr@rr©Ú __class__r6r7rDys  zAutoAugment.__init__)r@rAcCsJ|tjkrgd¢S|tjkr$gd¢S|tjkr6gd¢Std|›dƒ‚dS)N)))r'çš™™™™™Ù?é)r!ç333333ã?é ©)r(rLé©r)rLN©©r*çš™™™™™é?N©r*rLN))r'rLé)r'rLé©©r*rJN)r(çš™™™™™É?é)rX©r!rSrK))r(rLérT))r'rSrO©r*rN))r!rYr\)r(rLrK)rT)r'rJrV)r[©r$rJr))r!rJrMrT))r*rNrR©©r+rLNr]©)r$rLrZ)r%rrK)r[)r$ré))r$rSrK)r(rSrU))r&rJrUr`))rrLrOr])r^rTrWrNr_rarQ))©r+çš™™™™™¹?N)r%rYrV))r!çffffffæ?rb)rç333333Ó?rM))r&rSr)r&çÍÌÌÌÌÌì?r\))rçà?rK©r rerM))r)rhN©r*rgN))rrYrU)r'rfrU))r$rJr\)r#rLrU))r&rfrM)r#rerM)rT)r*rhN))r%rLrU)r&rLrO))r$rerU)rrhrK))r*rfN)r)rJN))r rJr\)r&rYrV))r#rgrV)r$rYrK))r(rhrb)r+rN)©r*rYNrP)rkrT))r$rgrMrT)©r)rSN)r(rYrK))r#rdr\)r$rer))r(rJrO©r)rgN))r rgrMri)rm)r(rSr\)rRrc)rirm))©rrgrZ)r+rYN)©rrgrK©r+reN)rT)r(rLrV©©r+rgNrT©rT)r!rgr\)rnrl)ro)r+rJN))rrgrO)r(rYrV)rrrlrs)rn)r(rfr\))rrSrKrp)rj)r rLrVrq))r%rfr\©r!rSrZ)©r+rSN)r rrb))rrerV)r(rJrK)r`rt))rrfrU)rrgr\))rrdrVr`))r(rerb)r rLrU))rrSrZru))rrerM)r rSr\))rrSrO)r)reN))rrerbrczThe provided policy r-)r r=r>r?r5)rGr@r6r6r7rE…s   zAutoAugment._get_policies©Únum_binsÚ image_sizerAcCsþt dd|¡dft dd|¡dft dd|d|¡dft dd|d|¡dft dd|¡dft dd|¡dft dd|¡dft dd|¡dft dd|¡dfd t |¡|dd  ¡ ¡d ft d d|¡d ft d¡d ft d¡d ft d¡d fd œS)NrrfTçtþÅÝ?rrç>@rgrKrZFçào@)rrrr r!r#r$r%r&r'r(r)r*r+)ÚtorchÚlinspaceÚarangeÚroundr2Útensor©rGrwrxr6r6r7Ú_augmentation_spaceßs$   ñzAutoAugment._augmentation_space)Ú transform_numrAcCs4tt |d¡ ¡ƒ}t d¡}t dd¡}|||fS)z†Get parameters for autoaugment transformation Returns: params required by the autoaugment transformation ©r)rbrb)r2r|ÚrandintÚitemZrand)rƒZ policy_idÚprobsÚsignsr6r6r7Ú get_paramsòs  zAutoAugment.get_params©rrAcCsü|j}t |¡\}}}t|tƒrTt|ttfƒr>t|ƒg|}n|durTdd„|Dƒ}| t|j ƒ¡\}}}|  d||f¡} t |j |ƒD]n\} \} } } || | krˆ| | \}}| durÆt||   ¡ƒnd}|râ|| dkrâ|d9}t || ||j|d}qˆ|S) z‘ img (PIL Image or Tensor): Image to be transformed. Returns: PIL Image or Tensor: AutoAugmented image. NcSsg|] }t|ƒ‘qSr6©Úfloat©Ú.0Úfr6r6r7Ú óz'AutoAugment.forward..é rrçð¿r")rr.Úget_dimensionsÚ isinstancerr2rŒr‰ÚlenrFr‚Ú enumerater†r8r)rGrrÚchannelsÚheightÚwidthZ transform_idr‡rˆÚop_metaÚirÚpZ magnitude_idÚ magnitudesÚsignedrr6r6r7Úforwardÿs"   zAutoAugment.forward©rAcCs|jj›d|j›d|j›dS)Nz(policy=ú, fill=ú))rIr9r@r)rGr6r6r7Ú__repr__szAutoAugment.__repr__)r9r:r;r<r r=r ÚNEARESTrrrŒrDrÚstrr2rErrÚboolr‚Ú staticmethodr‰r r¤Ú __classcell__r6r6rHr7r hs$ü û *þ Z* r csŽeZdZdZdddejdfeeeeeee ddœ‡fdd„ Z ee eefe e e eeffd œd d „Zeed œd d„Ze dœdd„Z‡ZS)r a~RandAugment data augmentation method based on `"RandAugment: Practical automated data augmentation with a reduced search space" `_. If the image is torch Tensor, it should be of type torch.uint8, and it is expected to have [..., 1 or 3, H, W] shape, where ... means an arbitrary number of leading dimensions. If img is PIL Image, it is expected to be in mode "L" or "RGB". Args: num_ops (int): Number of augmentation transformations to apply sequentially. magnitude (int): Magnitude for all the transformations. num_magnitude_bins (int): The number of different magnitude values. interpolation (InterpolationMode): Desired interpolation enum defined by :class:`torchvision.transforms.InterpolationMode`. Default is ``InterpolationMode.NEAREST``. If input is Tensor, only ``InterpolationMode.NEAREST``, ``InterpolationMode.BILINEAR`` are supported. fill (sequence or number, optional): Pixel fill value for the area outside the transformed image. If given a number, the value is used for all bands respectively. rbrMéN)Únum_opsrÚnum_magnitude_binsrrrAcs,tƒ ¡||_||_||_||_||_dSrB)rCrDr«rr¬rr)rGr«rr¬rrrHr6r7rD2s  zRandAugment.__init__rvcCsþt d¡dft dd|¡dft dd|¡dft dd|d|¡dft dd|d|¡dft dd|¡dft dd |¡dft dd |¡dft dd |¡dft dd |¡dfd t |¡|dd  ¡ ¡dft d d|¡dft d¡dft d¡dfd œS)NrFrfTryrrrzrgrKrZr{©r,rrrr r!r#r$r%r&r'r(r)r*©r|r€r}r~rr2rr6r6r7r‚As $  ñzRandAugment._augmentation_spacerŠc Csú|j}t |¡\}}}t|tƒrTt|ttfƒr>t|ƒg|}n|durTdd„|Dƒ}| |j||f¡}t |j ƒD]„}tt   t |ƒd¡ ¡ƒ}t| ¡ƒ|} || \} } | jdkrÄt| |j ¡ƒnd} | ràt   dd¡rà| d9} t|| | |j|d }qp|S) ú img (PIL Image or Tensor): Image to be transformed. Returns: PIL Image or Tensor: Transformed image. NcSsg|] }t|ƒ‘qSr6r‹rr6r6r7rar‘z'RandAugment.forward..r„rrrbr“r")rr.r”r•rr2rŒr‚r¬Úranger«r|r…r–r†ÚlistÚkeysÚndimrr8r) rGrrr˜r™ršr›Ú_Úop_indexrržrŸrr6r6r7r Ts"   zRandAugment.forwardr¡c Cs:|jj›d|j›d|j›d|j›d|j›d|j›d }|S)Nz (num_ops=z , magnitude=z, num_magnitude_bins=ú, interpolation=r¢r£)rIr9r«rr¬rr©rGÚsr6r6r7r¤os ÿþýüûÿ zRandAugment.__repr__)r9r:r;r<r r¥r2rrrŒrDrrr¦rr§r‚r r¤r©r6r6rHr7r s"ú ù*r cs|eZdZdZdejdfeeeee ddœ‡fdd„ Z ee e e eeffdœdd „Zeed œd d „Ze d œdd„Z‡ZS)raûDataset-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. rªN)r¬rrrAcs tƒ ¡||_||_||_dSrB)rCrDr¬rr)rGr¬rrrHr6r7rDŒs zTrivialAugmentWide.__init__)rwrAcCsît d¡dft dd|¡dft dd|¡dft dd|¡dft dd|¡dft dd|¡dft dd|¡dft dd|¡dft dd|¡dft dd|¡dfdt |¡|dd  ¡ ¡dft d d|¡dft d¡dft d¡dfd œS) NrFg®Gáz®ï?Tg@@gà`@rKrrVr{r­r®)rGrwr6r6r7r‚—s $  ñz&TrivialAugmentWide._augmentation_spacerŠc Csð|j}t |¡\}}}t|tƒrTt|ttfƒr>t|ƒg|}n|durTdd„|Dƒ}| |j¡}tt   t |ƒd¡  ¡ƒ}t | ¡ƒ|}||\} } | jdkrÀt| t j t | ƒdt jd  ¡ƒnd} | rÜt   dd¡rÜ| d 9} t||| |j|d S) r¯NcSsg|] }t|ƒ‘qSr6r‹rr6r6r7r·r‘z.TrivialAugmentWide.forward..r„r©Údtyperrbr“r")rr.r”r•rr2rŒr‚r¬r|r…r–r†r±r²r³Úlongr8r) rGrrr˜r™ršr›rµrržrŸrr6r6r7r ªs$   ÿ$ýzTrivialAugmentWide.forwardr¡cCs*|jj›d|j›d|j›d|j›d}|S)Nz(num_magnitude_bins=r¶r¢r£)rIr9r¬rrr·r6r6r7r¤Çs ÿþýÿzTrivialAugmentWide.__repr__)r9r:r;r<r r¥r2rrrŒrDrr¦rrr§r‚r r¤r©r6r6rHr7r|sü û rc sÒeZdZdZdddddejdfeeeeeee e eddœ‡fdd „ Z ee eefe ee eeffd œd d „Zejjed œdd„ƒZejjedœdd„ƒZeedœdd„Zeedœdd„Zed œdd„Z‡ZS)ra¼AugMix 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. r\éÿÿÿÿrTN)ÚseverityÚ mixture_widthÚ chain_depthÚalphaÚall_opsrrrAcsntƒ ¡d|_d|kr&|jks@ntd|j›d|›dƒ‚||_||_||_||_||_||_ ||_ dS)Nr’rz!The severity must be between [1, z]. Got z instead.) rCrDÚ_PARAMETER_MAXr5r½r¾r¿rÀrÁrr)rGr½r¾r¿rÀrÁrrrHr6r7rDçs zAugMix.__init__rvc Cs t dd|¡dft dd|¡dft d|dd|¡dft d|dd|¡dft dd|¡dfdt |¡|dd ¡ ¡d ft d d|¡d ft d¡d ft d¡d fd œ }|jr| t dd |¡dft dd |¡dft dd |¡dft dd |¡dfd œ¡|S)NrrfTrg@rrzrZFr{) rrrr r!r'r(r)r*rg)r#r$r%r&)r|r}r~rr2r€rÁÚupdate)rGrwrxr¸r6r6r7r‚ýs&$  ö üÿzAugMix._augmentation_spacer¡cCs t |¡SrB)r.Z pil_to_tensor©rGrr6r6r7Ú_pil_to_tensorszAugMix._pil_to_tensor)rcCs t |¡SrB)r.Z to_pil_imagerÄr6r6r7Ú_tensor_to_pilszAugMix._tensor_to_pil)ÚparamsrAcCs t |¡SrB)r|Ú_sample_dirichlet)rGrÇr6r6r7rÈszAugMix._sample_dirichlet)Úorig_imgrAc Cs||j}t |¡\}}}t|tƒrZ|}t|ttfƒrBt|ƒg|}qd|durddd„|Dƒ}n | |¡}| |j ||f¡}t |j ƒ}|  dgt d|jdƒ|¡} |  d¡gdg| jd} | tj|j|jg| jd | dd¡¡} | tj|jg|j| jd | dd¡¡| dd…df  | ddg¡} | dd…df  | ¡| } t|jƒD]ø}| }|jdkrn|jnttjddd d  ¡ƒ}t|ƒD]š}tt t|ƒd ¡ ¡ƒ}t | ¡ƒ|}||\}}|jdkròt|tj|jd tjd  ¡ƒnd }|rt d d ¡r|d9}t||||j |d}qŽ|  !| dd…|f  | ¡|¡qT|   |¡j"|j#d } t|tƒsx| $| ¡S| S)r¯NcSsg|] }t|ƒ‘qSr6r‹rr6r6r7r/r‘z"AugMix.forward..rrZr)Údevicer¼r„)ÚlowÚhighÚsizer¹rrbr“r")%rr.r”r•rr2rŒrÅr‚rÂr±ÚshapeÚviewÚmaxr³rÍrÈr|r€rÀrÊÚexpandr¾r°r¿r…r†r–r²r½r»r8rZadd_ÚtorºrÆ)rGrÉrr˜r™ršrr›Z orig_dimsÚbatchZ batch_dimsÚmZcombined_weightsZmixrœÚaugÚdepthr´rµrržrŸrr6r6r7r !sR    "ÿ$ÿþ*  ÿ$ý$  zAugMix.forwardcCsJ|jj›d|j›d|j›d|j›d|j›d|j›d|j›d|j›d}|S) Nz (severity=z, mixture_width=z, chain_depth=z, alpha=z , all_ops=r¶r¢r£) rIr9r½r¾r¿rÀrÁrrr·r6r6r7r¤[s" ÿþýüûúùÿ zAugMix.__repr__)r9r:r;r<r ZBILINEARr2rŒr§rrrDrrr¦rr‚r|ZjitZunusedrÅrÆrÈr r¤r©r6r6rHr7rÒs4ø ÷*:r)r/ÚenumrÚtypingrrrrr|rÚr r.r Ú__all__r¦rŒr8r ÚnnÚModuler r rrr6r6r6r7Ús  ÿ P 8]V