Ei@n SSKrSSKJr SSKJr SSKrSSKJr SSKJr J r /SQr S\S \ S \ S \ S \\\ 4 S jr"SS\5r"SS\R"R$5r"SS\R"R$5r"SS\R"R$5r"SS\R"R$5rg)N)Enum)Optional)Tensor) functionalInterpolationMode)AutoAugmentPolicy AutoAugment RandAugmentTrivialAugmentWideAugMiximgop_name magnitude interpolationfillc US:XaK[R"USSS/S[R"[R"U55S/UUSS/S9nU$US:XaK[R"USSS/SS[R"[R"U55/UUSS/S9nU$US:Xa)[R"US[ U5S/SUSS/US9nU$US :Xa)[R"USS[ U5/SUSS/US9nU$US :Xa[R "XX4S 9nU$US :Xa[R"USU-5nU$US :Xa[R"USU-5nU$US:Xa[R"USU-5nU$US:Xa[R"USU-5nU$US:Xa"[R"U[ U55nU$US:Xa[R"X5nU$US:Xa[R"U5nU$US:Xa[R"U5nU$US:Xa[R"U5nU$US:XaU$[!SUS35e)NShearXr?)angle translatescaleshearrrcenterShearY TranslateX)rrrrrr TranslateYRotaterr BrightnessColorContrast Sharpness PosterizeSolarize AutoContrastEqualizeInvertIdentityzThe provided operator  is not recognized.)Faffinemathdegreesatanintrotateadjust_brightnessadjust_saturationadjust_contrastadjust_sharpness posterizesolarize autocontrastequalizeinvert ValueError)rrrrrs _/home/mande/repo/quber/.venv/lib/python3.13/site-packages/torchvision/transforms/autoaugment.py _apply_opr> s(hh !f<< ) 45s;'q6  F Js H hh !f TYYy%9:;'q6  l JY L hh 9~q)'* V JE L hh #i.)'* B J1 H hhs]N. J- L !!#sY7* J) G !!#sY7& J% J S9_5" J! K   cIo6 J K kk#s9~. J J jj( J N "nnS! J J jjo J H hhsm J J   J1':MNOOc$\rSrSrSrSrSrSrSrg)r ]zgAutoAugment policies learned on different datasets. Available policies are IMAGENET, CIFAR10 and SVHN. imagenetcifar10svhnN) __name__ __module__ __qualname____firstlineno____doc__IMAGENETCIFAR10SVHN__static_attributes__rEr?r=r r ]sHG Dr?r c @^\rSrSrSr\R \RS4S\S\S\ \ \ SS4U4Sjjjr S\S\ \ \ \\ \ \4\ \\ \ \444S jrS \S \ \\4S\\\ \\444S jr\S \S\ \\\44Sj5rS\S\4SjrS\4SjrSrU=r$)r ha 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. Npolicyrrreturncr>[TU]5 XlX lX0lUR U5UlgN)super__init__rQrr _get_policiespolicies)selfrQrr __class__s r=rVAutoAugment.__init__ys2  * **62 r?cU[R:Xa/SQ$U[R:Xa/SQ$U[R:Xa/SQ$[ SUS35e)N)))r%皙?)r333333? )r&r_r'r_Nr(皙?Nr(r_N))r%r_)r%r_r(r]N)r&皙?)rkrrfr^))r&r_rg))r%rfrbr(rN))rrlro)r&r_r^)rg)r%r]ri)rnr"r]r))rr]r`rg))r(rNrer)r_Nrp)r"r_rm)r#rr^)rn)r"r))r"rfr^)r&rfrh))r$r]rhrs))rr_rbrp)rqrgrjrarrrtrd))r)皙?N)r#rlri))rffffff?ru)r333333?r`))r$rfr)r$?ro))r?r^rrxr`))r'r{Nr(rzN))rrlrh)r%ryrh))r"r]ro)r!r_rh))r$ryr`)r!rxr`)rg)r(r{N))r#r_rh)r$r_rb))r"rxrh)rr{r^))r(ryN)r'r]N))rr]ro)r$rlri))r!rzri)r"rlr^))r&r{ru)r)rN)r(rlNrc)r~rg))r"rzr`rg)r'rfN)r&rlr^))r!rwro)r"rxr))r&r]rbr'rzN))rrzr`r|)r)r&rfro)rerv)r|r))rrzrm)r)rlN)rrzr^r)rxN)rg)r&r_rir)rzNrgrg)rrzro)rr)r)r)r]N))rrzrb)r&rlri)rrr)r)r&ryro))rrfr^r)r})rr_rir))r#ryrorrfrm)r)rfN)rrru))rrxri)r&r]r^)rsr))rryrh)rrzro))rrwrirs))r&rxru)rr_rh))rrfrmr))rrxr`)rrfro))rrfrb)r'rxN))rrxrurvzThe provided policy r+)r rKrLrMr<)rYrQs r=rWAutoAugment._get_policiessk &// / 6(00 0 6(-- - 83F8;NOP Pr?num_bins image_sizec[R"SSU5S4[R"SSU5S4[R"SSUS-U5S4[R"SSUS-U5S4[R"SSU5S4[R"SSU5S4[R"SSU5S4[R"SSU5S4[R"SSU5S4S [R"U5US- S - - R5R 5- S 4[R"S SU5S 4[R "S5S 4[R "S5S 4[R "S5S 4S .$)NrryTt?rr>@rzr^rmFo@)rrrrrr!r"r#r$r%r&r'r(r))torchlinspacearangeroundr1tensorrYrrs r=_augmentation_spaceAutoAugment._augmentation_spaces^~~c394@~~c394@ >>#}z!}/LhWY]^ >>#}z!}/LhWY]^~~c4:DA >>#sH=tDnnS#x8$?S(;TB..c8r)rYrrchannelsheightwidthf transform_idrrop_metairp magnitude_id magnitudessignedrs r=forwardAutoAugment.forwards-yy"#"2"23"7% c6 " "$e --d }x/!*./$Qa$/%)__S5G%H" U**2?-6t}}\7R-S )A)Lx1}%,%5" FFRF^E*":"?"?"ABdg eh!m%IitGYGY`de .T 0s-EchURRSURSURS3$)Nz(policy=, fill=))rZrFrQr)rYs r=__repr__AutoAugment.__repr__s/..))*(4;;-wtyykQRSSr?)rrrXrQ)rFrGrHrIrJr rKrNEARESTrlistrrVtuplestrr1rWdictrboolr staticmethodrrrrN __classcell__rZs@r=r r hs4$%6$>$>+<+D+D&* 3! 3) 3tE{# 3  3 3XQ'XQ eE#uhsm34eCQT `_. If the image is torch Tensor, it should be of type torch.uint8, and it is expected to have [..., 1 or 3, H, W] shape, where ... means an arbitrary number of leading dimensions. If img is PIL Image, it is expected to be in mode "L" or "RGB". Args: num_ops (int): Number of augmentation transformations to apply sequentially. magnitude (int): Magnitude for all the transformations. num_magnitude_bins (int): The number of different magnitude values. interpolation (InterpolationMode): Desired interpolation enum defined by :class:`torchvision.transforms.InterpolationMode`. Default is ``InterpolationMode.NEAREST``. If input is Tensor, only ``InterpolationMode.NEAREST``, ``InterpolationMode.BILINEAR`` are supported. fill (sequence or number, optional): Pixel fill value for the area outside the transformed image. If given a number, the value is used for all bands respectively. rur`Nnum_opsrnum_magnitude_binsrrrRc^>[TU]5 XlX lX0lX@lXPlgrT)rUrVrrrrr)rYrrrrrrZs r=rVRandAugment.__init__2s+  ""4* r?rrc[R"S5S4[R"SSU5S4[R"SSU5S4[R"SSUS-U5S4[R"SSUS-U5S4[R"SSU5S4[R"SS U5S4[R"SS U5S4[R"SS U5S4[R"SS U5S4S [R"U5US- S - - R 5R 5- S4[R"S SU5S4[R"S5S4[R"S5S4S .$)NrFryTrrrrrzr^rmrr*rrrrrr!r"r#r$r%r&r'r(rrrrrr1rs r=rRandAugment._augmentation_spaceAs^c*E2~~c394@~~c394@ >>#}z!}/LhWY]^ >>#}z!}/LhWY]^~~c4:DA >>#sH=tDnnS#x8$?S(;TB..c8r)rYrrrrrrr_op_indexrrrrs r=rRandAugment.forwardTsDyy"#"2"23"7% c6 " "$e --d }x/!*./$Qa$/**4+B+BVOTt||$A5==Wt<AACDH7<<>*84G!(!1 JDNOOVWDWj8==?@]`I%--400T! C)CUCU\`aC% 0s-F c URRSURSURSURSUR SUR S3 nU$)Nz (num_ops=z , magnitude=z, num_magnitude_bins=, interpolation=rr)rZrFrrrrrrYss r=rRandAugment.__repr__osg~~&&'(||n4>>*#D$;$;#<t112dii[   r?)rrrrr)rFrGrHrIrJrrr1rrrrVrrrrrrrrrNrrs@r=r r s("$+<+D+D&*     )  tE{#     C U38_ QUVY[`agimam[nVnQo &6f6 #  r?r c ^\rSrSrSrS\R S4S\S\S\\ \ SS4U4S jjjr S \S\ \ \\\444S jrS \S\4S jrS\ 4SjrSrU=r$)r i|aDataset-independent data-augmentation with TrivialAugment Wide, as described in `"TrivialAugment: Tuning-free Yet State-of-the-Art Data Augmentation" `_. If the image is torch Tensor, it should be of type torch.uint8, and it is expected to have [..., 1 or 3, H, W] shape, where ... means an arbitrary number of leading dimensions. If img is PIL Image, it is expected to be in mode "L" or "RGB". Args: num_magnitude_bins (int): The number of different magnitude values. interpolation (InterpolationMode): Desired interpolation enum defined by :class:`torchvision.transforms.InterpolationMode`. Default is ``InterpolationMode.NEAREST``. If input is Tensor, only ``InterpolationMode.NEAREST``, ``InterpolationMode.BILINEAR`` are supported. fill (sequence or number, optional): Pixel fill value for the area outside the transformed image. If given a number, the value is used for all bands respectively. rNrrrrRcF>[TU]5 XlX lX0lgrT)rUrVrrr)rYrrrrZs r=rVTrivialAugmentWide.__init__s! "4* r?rc[R"S5S4[R"SSU5S4[R"SSU5S4[R"SSU5S4[R"SSU5S4[R"SSU5S4[R"SSU5S4[R"SSU5S4[R"SSU5S4[R"SSU5S4S[R"U5US- S - - R 5R 5- S4[R"S SU5S4[R"S5S4[R"S5S4S .$) NrFgGz?Tg@@g`@r^rrirrr)rYrs r=r&TrivialAugmentWide._augmentation_spacesJc*E2~~c4:DA~~c4:DA >>#tX>E >>#tX>E~~c5(;TB >>#tX>EnnS$94@T8r) rYrrrrrrrrrrrrs r=rTrivialAugmentWide.forwardsCyy"#"2"23"7% c6 " "$e --d }x/!*./$Qa$/**4+B+BCu}}S\48==?@w||~&x0$- " *U]]3z?D STYY[ \  emmAt,,  Iy@R@RY]^^0s-FcURRSURSURSURS3nU$)Nz(num_magnitude_bins=rrr)rZrFrrrrs r=rTrivialAugmentWide.__repr__sP~~&&'(""&"9"9!:t112dii[   r?)rrr)rFrGrHrIrJrrr1rrrrVrrrrrrrrrNrrs@r=r r |s "#%+<+D+D&*  ) tE{#     C DeFDL>Q9Q4R &_6_f_:#r?r ch^\rSrSrSrSSSSS\R S4S\S \S \S \S \ S \S\ \ \SS4U4Sjjjr S\S\ \\4S\\\ \\ 444Sjr\R&R(S\4Sj5r\R&R(S\4Sj5rS\S\4SjrS\S\4SjrS\4SjrSrU=r$)r iaxAugMix 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. rorTNseverity mixture_width chain_depthalphaall_opsrrrRc>[TU]5 SUlSUs=::aUR::dO [SURSUS35eXlX lX0lX@lXPlX`l Xpl g)Nrrz!The severity must be between [1, z]. Got z instead.) rUrV_PARAMETER_MAXr<rrrrrrr) rYrrrrrrrrZs r=rVAugMix.__init__sw  X4!4!44@ATAT@UU\]e\ffopq q *&  * r?rrc 8[R"SSU5S4[R"SSU5S4[R"SUSS- U5S4[R"SUSS- U5S4[R"SSU5S4S[R"U5US- S- - R5R 5- S 4[R"S SU5S 4[R "S5S 4[R "S5S 4S . nUR (avUR[R"SS U5S4[R"SS U5S4[R"SS U5S4[R"SS U5S4S .5 U$)NrryTrg@rrrmFr) rrrrrr%r&r'r(rz)r!r"r#r$)rrrrr1rrupdate)rYrrrs r=rAugMix._augmentation_spacesi~~c394@~~c394@ >>#z!}s/BHMtT >>#z!}s/BHMtT~~c4:DAu||H5(Q,!9KLSSUYY[[]bcsH=uE"\\#.6c*E2  << HH#(>>#sH#Et"L#nnS#x@$G!&S(!CT J"'..c8"Dd!K  r?c.[R"U5$rT)r, pil_to_tensorrYrs r=_pil_to_tensorAugMix._pil_to_tensorss##r?rc.[R"U5$rT)r, to_pil_imagers r=_tensor_to_pilAugMix._tensor_to_pils~~c""r?paramsc.[R"U5$rT)r_sample_dirichlet)rYrs r=rAugMix._sample_dirichlets&&v..r?orig_imgc URn[R"U5up4n[U[5(aLUn[U[ [ 45(a[ U5/U-nO0UbUVs/sHn[ U5PM nnOURU5nURURXE45n[UR5n URS/[SUR- S5-U -5n U RS5/S/U RS- --n UR!["R$"UR&UR&/U R(S9R+U SS55n UR!["R$"UR&/UR,-U R(S9R+U SS55U SS2S4RU SS/5-n U SS2S4RU 5U -n[/UR,5GHnU nUR0S:a UR0O,[ ["R2"SSSS9R555n[/U5Hn[ ["R2"[7U5S5R555n[UR955UnUUunnURS:aH[ U["R2"UR:S["R<S 9R555OS nU(a!["R2"S S5(aUS -n[?UUUUR@US 9nM URCU SS2U4RU 5U-5 GM URU 5REURFS 9n[U[5(dURIU5$U$s snf)rNrrmr)devicerr)lowhighsizerrrurr )%rr,rrrr1rrrrrshapeviewmaxrrrrrrrexpandrrrrrrrrrr>radd_torr)rYrrrrrrrr orig_dimsbatch batch_dimsmcombined_weightsmixraugdepthrrrrrrs r=rAugMix.forward!sYyy"#"2"28"<% h ' 'C$e --d }x/!*./$Qa$/%%h/C**4+>+>PO !s1sxx<33i?@jjm_sejj1n'==   " " LL$**djj1%,, G N NzZ[}^` a   11 LL$**(:(::5<< P W WXbcdXegi j adGLL*Q-, -.1gll:&.t))*AC(,(8(81(r nnModuler r r r rEr?r=r#s  0 ]M MM*/M@QMYabfglbmYnM`tT%((//tTnZ%((//ZzSSlUUXX__Ur?