ڧgSn nddlZddlmZddlmZmZmZmZddlZddlm Z ddl m Z m Z gdZde d ed ed e d eeef d ZGddeZGddejjZGddejjZGddejjZGddejjZdS)N)Enum)DictListOptionalTuple)Tensor) functionalInterpolationMode)AutoAugmentPolicy AutoAugment RandAugmentTrivialAugmentWideAugMiximgop_name magnitude interpolationfillc |dkrHtj|dddgdtjtj|dg||ddg}n|dkrHtj|dddgddtjtj|g||ddg}n|dkr.tj|dt |dgd|ddg|}n|d kr.tj|ddt |gd|ddg|}nM|d krtj|||| }n-|d krtj|d|z}n |d krtj|d|z}n|dkrtj |d|z}n|dkrtj |d|z}n|dkr#tj |t |}n|dkrtj ||}nk|dkrtj |}nP|dkrtj|}n5|dkrtj|}n|dkrnt!d|d|S)NShearXr?)angle translatescaleshearrrcenterShearY TranslateX)rrrrrr TranslateYRotaterr BrightnessColorContrast Sharpness PosterizeSolarize AutoContrastEqualizeInvertIdentityzThe provided operator  is not recognized.)Faffinemathdegreesatanintrotateadjust_brightnessadjust_saturationadjust_contrastadjust_sharpness posterizesolarize autocontrastequalizeinvert ValueError)rrrrrs ^/var/www/html/ai-engine/env/lib/python3.11/site-packages/torchvision/transforms/autoaugment.py _apply_oprA s(h !f< ) 4 455s;'q6     H  h !f TYy%9%9::;'q6     L h 9~~q)'*    L h #i..)'*    H  hsI]NNN L !#sY77 G  !#sY77 J  S9_55 K   cIo66 K  k#s9~~.. J  ji(( N " "nS!! J  joo H  hsmm J   N'NNNOOO JceZdZdZdZdZdZdS)r zoAutoAugment policies learned on different datasets. Available policies are IMAGENET, CIFAR10 and SVHN. imagenetcifar10svhnN)__name__ __module__ __qualname____doc__IMAGENETCIFAR10SVHNrBr@r r ]s)HG DDDrBr c beZdZdZejejdfdededee e ddffd Z dede e e e e eefe e e eefffdZd ed e eefdee e eefffd Zed ede eeeffd ZdedefdZde fdZxZS)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. Npolicyrrreturnct||_||_||_|||_dSN)super__init__rPrr _get_policiespolicies)selfrPrr __class__s r@rUzAutoAugment.__init__ysJ  * **622 rBc|tjkrgdS|tjkrgdS|tjkrgdSt d|d)N)))r(皙?)r"333333? )r)r]r*r]Nr+皙?Nr+r]N))r(r])r(r]r+r[N)r)皙?)rir"rdr\))r)r]re))r(rdr`r+rN))r"rjrm)r)r]r\)re)r(r[rg)rlr%r[r))r"r[r^re))r+rNrcr,r]Nrn)r%r]rk)r&rr\)rl)r%r))r%rdr\)r)rdrf))r'r[rfrq))rr]r`rn)rorerhr_rprrrb))r,皙?N)r&rjrg))r"ffffff?rs)r 333333?r^))r'rdr )r'?rm))r?r\r!rvr^))r*ryNr+rxN))rrjrf)r(rwrf))r%r[rm)r$r]rf))r'rwr^)r$rvr^)re)r+ryN))r&r]rf)r'r]r`))r%rvrf)r ryr\))r+rwN)r*r[N))r!r[rm)r'rjrg))r$rxrg)r%rjr\))r)ryrs)r,rN)r+rjNra)r|re))r%rxr^re)r*rdN)r)rjr\))r$rurm)r%rvr))r)r[r`r*rxN))r!rxr^rz)r~)r)rdrm)rcrt)rzr~))rrxrk)r,rjN)rrxr\r,rvN)re)r)r]rgr,rxNrere)r"rxrm)rr})r)r,r[N))rrxr`)r)rjrg)rr}r)r)r)rwrm))rrdr\r)r{)r!r]rgr))r&rwrmr"rdrk)r,rdN)r!rrs))rrvrg)r)r[r\)rqr))rrwrf)r rxrm))rrurgrq))r)rvrs)r!r]rf))rrdrkr))rrvr^)r!rdrm))rrdr`)r*rvN))rrvrsrtzThe provided policy r.)r rKrLrMr?)rXrPs r@rVzAutoAugment._get_policiess &/ / / 6(0 0 0 6(- - - 8OFOOOPP PrBnum_bins image_sizectjdd|dftjdd|dftjdd|dz|dftjdd|dz|dftjdd|dftjdd|dftjdd|dftjdd|dftjdd|dfd tj||dz d z z z d ftjd d|d ftjdd ftjdd ftjdd fd S)NrrwTt?r r>@rxr\rkFo@)rrr r!r"r$r%r&r'r(r)r*r+r,)torchlinspacearangeroundr4tensorrXrrs r@_augmentation_spacezAutoAugment._augmentation_spacesw~c3994@~c3994@ >#}z!}/LhWWY]^ >#}z!}/LhWWY]^~c4::DA >#sH==tDnS#x88$?S(;;TB.c8<t|t t frt |g|z}n| d|D}|t|j \}}}| d||f} t|j |D]w\} \} } } || | krb| | \}}| 't ||  nd}|r|| dkr|dz}t|| ||j|}x|S)z img (PIL Image or Tensor): Image to be transformed. Returns: PIL Image or Tensor: AutoAugmented image. Nc,g|]}t|SrNfloat.0fs r@ z'AutoAugment.forward.. ///Qa///rB rrr#)rr/get_dimensions isinstancerr4rrlenrWr enumeraterrAr)rXrrchannelsheightwidth transform_idrrop_metairp magnitude_id magnitudessignedrs r@forwardzAutoAugment.forwardscy"#"23"7"7&% c6 " " 0$e -- 0d }x/!//$///%)__S5G5G%H%H" eU**2??-6t}\7R-S-S f f )A)LQx1}}%,W%5" FFRF^E*\":"?"?"A"ABBBdg &eAh!mm%IWitGY`deee rBc@|jjd|jd|jdS)Nz(policy=, fill=))rYrGrPr)rXs r@__repr__zAutoAugment.__repr__s*.)SS4;SStySSSSrB)rGrHrIrJr rKr NEARESTrrrrUrstrr4rVrrboolr staticmethodrrr __classcell__rYs@r@r r hs$%6$>+<+D&* 3 3! 3) 3tE{# 3  3 3 3 3 3 3XQ'XQ eE#uhsm34eCQT `_. 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. rsr^Nnum_opsrnum_magnitude_binsrrrQct||_||_||_||_||_dSrS)rTrUrrrrr)rXrrrrrrYs r@rUzRandAugment.__init__2sD  ""4* rBrrctjddftjdd|dftjdd|dftjdd|dz|dftjdd|dz|dftjdd|dftjdd |dftjdd |dftjdd |dftjdd |dfd tj||dz d z z z dftjd d|dftjddftjddfd S)NrFrwTrr rrrxr\rkrr-rrr r!r"r$r%r&r'r(r)r*r+rrrrrr4rs r@rzRandAugment._augmentation_spaceAswc**E2~c3994@~c3994@ >#}z!}/LhWWY]^ >#}z!}/LhWWY]^~c4::DA >#sH==tDnS#x88$?S(;;TB.c8<t|t t frt |g|z}n| d|D}||j||f}t|j D]}t tj t|d}t||} || \} } | jdkr,t | |jnd} | rtj ddr| dz} t'|| | |j|}|S) img (PIL Image or Tensor): Image to be transformed. Returns: PIL Image or Tensor: Transformed image. Nc,g|]}t|SrNrrs r@rz'RandAugment.forward..arrBrrrrsrr#)rr/rrrr4rrrrangerrrrrlistkeysndimrrAr) rXrrrrrr_op_indexrrrrs r@rzRandAugment.forwardTs}y"#"23"7"7&% c6 " " 0$e -- 0d }x/!//$///**4+BVUOTTt|$$ b bA5=Wt<<AACCDDH7<<>>**84G!(!1 JDNOVWDWDWj8==??@@@]`I "%-400 "T! C)4CU\`aaaCC rBc t|jjd|jd|jd|jd|jd|jd }|S)Nz (num_ops=z , magnitude=z, num_magnitude_bins=, interpolation=rr)rYrGrrrrrrXss r@rzRandAugment.__repr__osx~&  |  >  %)$;   $1    i     rB)rGrHrIrJr rr4rrrrUrrrrrrrrrrs@r@rrs*("$+<+D&*      )  tE{#         C U38_ QUVY[`agimam[nVnQo    &6f6 #        rBrc eZdZdZdejdfdededeee ddffd Z d ede e e eefffd Zd edefd Zde fd ZxZS)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. rNrrrrQcrt||_||_||_dSrS)rTrUrrr)rXrrrrYs r@rUzTrivialAugmentWide.__init__s6 "4* rBrctjddftjdd|dftjdd|dftjdd|dftjdd|dftjdd|dftjdd|dftjdd|dftjdd|dftjdd|dfdtj||dz d z z z dftjd d|dftjddftjddfd S) NrFgGz?Tg@@g`@r\r rgrrr)rXrs r@rz&TrivialAugmentWide._augmentation_spacescc**E2~c4::DA~c4::DA >#tX>>E >#tX>>E~c5(;;TB >#tX>>EnS$994@T8<t|t t frt |g|z}n| d|D}||j}t tj t|d }t||}||\} } | jdkrSt | tj t| dtj nd} | rtj ddr| dz} t#||| |j| S) rNc,g|]}t|SrNrrs r@rz.TrivialAugmentWide.forward..rrBrrdtyperrsrr#)rr/rrrr4rrrrrrrrrrlongrAr) rXrrrrrrrrrrrs r@rzTrivialAugmentWide.forwardsvy"#"23"7"7&% c6 " " 0$e -- 0d }x/!//$///**4+BCCu}S\\488==??@@w||~~&&x0$W- F"" *U]3z??D SSSTYY[[ \ \ \   emAt,,   Igy@RY]^^^^rBcT|jjd|jd|jd|jd}|S)Nz(num_magnitude_bins=rrr)rYrGrrrrs r@rzTrivialAugmentWide.__repr__sV~&  "&"9  #1  i    rB)rGrHrIrJr rr4rrrrUrrrrrrrrrrs@r@rr|s  "#%+<+D&*   ) tE{#         C DeFDL>Q9Q4R    &_6_f____:#rBrcFeZdZdZdddddejdfdeded ed ed ed ed e e eddffd Z dede eefde ee eefffdZejjdefdZejjdefdZdedefdZdedefdZdefdZxZS)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. rmrTNseverity mixture_width chain_depthalphaall_opsrrrQctd|_d|cxkr |jksntd|jd|d||_||_||_||_||_||_ ||_ dS)Nrr z!The severity must be between [1, z]. Got z instead.) rTrU_PARAMETER_MAXr?rrrrrrr) rXrrrrrrrrYs r@rUzAugMix.__init__s  X4444!44444pATpp]epppqq q  *&  * rBrrc tjdd|dftjdd|dftjd|ddz |dftjd|ddz |dftjdd|dfdtj||dz dz z z d ftjd d|d ftjdd ftjdd fd }|jrr|tjdd |dftjdd |dftjdd |dftjdd |dfd |S)NrrwTr g@rrrkFr) rrr r!r"r(r)r*r+rx)r$r%r&r')rrrrr4rrupdate)rXrrrs r@rzAugMix._augmentation_spaces~c3994@~c3994@ >#z!}s/BHMMtT >#z!}s/BHMMtT~c4::DAu|H55(Q,!9KLSSUUYY[[[]bcsH==uE"\#..6c**E2  <  HH#(>#sH#E#Et"L#nS#x@@$G!&S(!C!CT J"'.c8"D"Dd!K    rBc*tj|SrS)r/ pil_to_tensorrXrs r@_pil_to_tensorzAugMix._pil_to_tensorss###rBrc*tj|SrS)r/ to_pil_imagers r@_tensor_to_pilzAugMix._tensor_to_pils~c"""rBparamsc*tj|SrS)r_sample_dirichlet)rXrs r@rzAugMix._sample_dirichlets&v...rBorig_imgc |j}tj|\}}}t|trA|}t|t t frt |g|z}n$| d|D}n||}||j ||f}t|j }| dgtd|jz dz|z} | dgdg| jdz zz} |t#j|j|jg| j| dd} |t#j|jg|jz| j| dd| dddf | ddgz} | dddf | | z} t/|jD]}| }|jdkr|jn5t t#jddd }t/|D]}t t#jt7|d}t||}||\}}|jdkrKt |t#j|jdt"j nd }|rt#jd dr|d z}t?||||j |}| !| dd|f | |z| |"|j# } t|ts|$| S| S)rNc,g|]}t|SrNrrs r@rz"AugMix.forward../rrBr rkr)devicerr)lowhighsizerrrsrr#)%rr/rrrr4rrrrrshapeviewmaxrrrrrrrexpandrrrrrrrrrrAradd_torr)rXrrrrrrr orig_dimsbatch batch_dimsmcombined_weightsmixraugdepthrrrrrrs r@rzAugMix.forward!sy"#"28"<"<&% h ' ' 0C$e -- 0d }x/!//$///%%h//C**4+>PPOO !s1sx<333i?@@jjmm_sej1n'==   " " L$*dj1%, G G G N NzZ[}^` a a    11 L$*(::5< P P P W WXbcdXegi j j  aaadGLL*Q-, - -.1gll:&&.t)** D DAC(,(81(<(rs ............ 00000000 ] ] ]M MM*/M@QMYabfglbmYnMMMM`tTtTtTtTtT%(/tTtTtTnZZZZZ%(/ZZZzSSSSSSSSlUUUUUUX_UUUUUrB