Ph~ddlZddlmZmZmZmZmZmZmZm Z ddl Z ddl Z ddl mZmZmZddlmZmZddlmZddlmZmZmZmZddlmZddl m!Z!dd l"m#Z#m$Z$d d l%m&Z&m'Z'm(Z(m)Z)e e jTe jVjVejVejXfZ-Gd d eZ.Gdde.Z/Gdde.Z0Gdde.Z1Gdde.Z2y)N)AnyCallableDictListOptionalTupleTypeUnion) tree_flattentree_unflattenTreeSpec) transforms tv_tensors)_functional_tensor)AutoAugmentPolicy functionalInterpolationMode Transform)_check_interpolation)get_size) _FillType _FillTypeJIT) _get_fill_setup_fill_arg check_typeis_pure_tensorc eZdZejdddeeefdeeeee e feffddffdZ dee e fffd Z dee eeeffdee eeefffd Zej&ej(ffd e d ee d fdeeee eefeffd Zdeee eefdede fdZdede dedeeefdeee e fefdef dZxZS)_AutoAugmentBaseN interpolationfillr!r"returncpt|t||_||_t ||_yN)super__init__rr!r"r_fill)selfr!r" __class__s rC:\Users\daisl\Desktop\realtime-object-detection\venv\Lib\site-packages\torchvision/transforms/v2/_auto_augment.pyr'z_AutoAugmentBase.__init__s0 1-@ $T* ct|}t|dtr!t t |j d|S)Nr"z7() can not be scripted for when `fill` is a dictionary.)r& _extract_params_for_v1_transform isinstancedict ValueErrortype__name__)r)paramsr*s r+r.z1_AutoAugmentBase._extract_params_for_v1_transform!sA9; fVnd +T 3 344klm m r,dctct|j}|ttjt |d}|||fS)N)tuplekeysinttorchrandintlen)r)r5r9keys r+_get_random_itemz!_AutoAugmentBase._get_random_item)s>SXXZ 3u}}SY345CH}r,inputsunsupported_types.c rtt|dkDr|n|d\}}|j|}g}tt ||D]\}\}} | r\t |t jtjjtt jfr|j||fgt||sttdt|jdt|jd|s tdt|dkDrtdt|d|d\}} |||f| fS) NrrzInputs of type z are not supported by z()zFound no image in the sample.z^Auto augment transformations are only properly defined for a single image or video, but found .)r r=_needs_transform_list enumerateziprrImagePILrVideoappendr/ TypeErrorr2r3) r)r@rA flat_inputsspecneeds_transform_listimage_or_videosidxinptneeds_transformimage_or_videos r+#_flatten_and_extract_image_or_videoz4_AutoAugmentBase._flatten_and_extract_image_or_video.sF )3v;?q R T#99+F,5c+G[6\,] (C($:$$IIOO"$$ $ &&T{3D"34/$t*2E2E1FF\]abf]g]p]p\qqs tuu-^;< <  ! # 12!5  .a0^T3'77r,flat_inputs_with_specrSc0|\}}}|||<t||Sr%)r )r)rUrSrLrMrPs r+$_unflatten_and_insert_image_or_videoz5_AutoAugmentBase._unflatten_and_insert_image_or_videoPs' "7 T3) Ck400r,image transform_id magnitudec t|t|}|dk(r|S|dk(rItj|dddgdt j t j |dg||ddgS|dk(rItj|dddgddt j t j |g||ddgS|dk(r)tj|dt|dgd|ddg| S|d k(r)tj|ddt|gd|ddg| S|d k(rtj|||| S|d k(rtj|d|zS|dk(rtj|d|zS|dk(rtj|d|zS|dk(rtj|d|zS|dk(r tj|t|S|dk(rUt|tj rt#j$|j&nd}tj(|||zS|dk(rtj*|S|dk(rtj,|S|dk(rtj.|St1d|)NIdentityShearXr?)angle translatescaleshearr!r"centerShearY TranslateX)r`rarbr!rcr" TranslateYRotate)r`r!r" Brightness)brightness_factorColor)saturation_factorContrast)contrast_factor Sharpness)sharpness_factor Posterize)bitsSolarizego@) threshold AutoContrastEqualizeInvertzNo transform available for )rr2Faffinemathdegreesatanr:rotateadjust_brightnessadjust_saturationadjust_contrastadjust_sharpness posterizer/r;Tensor_FT _max_valuedtypesolarize autocontrastequalizeinvertr1)r)rXrYrZr!r"fill_bounds r+_apply_image_or_video_transformz0_AutoAugmentBase._apply_image_or_video_transformYs$U , : %L X %88a&||DIIi$893?+1v  X %88a&DLL9)=>?+1v  \ )88y>1-+Cj \ )88c)n-+Cj X %88E-V[\ \ \ )&&uiP P W $&&uiP P Z '$$UC)OL L [ (%%ecIoN N [ (;;u3y>: : Z '3=eU\\3RCNN5;;/X]E::euy/@A A ^ +>>%( ( Z '::e$ $ X %88E? ":<.IJ Jr,)r3 __module__ __qualname__rNEARESTr r:rrr strr'rr.rrboolr?r BoundingBoxesMaskrr ImageOrVideorTrWfloatrr __classcell__r*s@r+rrs8I7P7PDH +.34 +ItE$)$4i$?@@A +  +$sCx.DeHdN.C)C$DsTYZbdhZhTiOiIj0:/G/G.Y 8 8!s+ 8 uT#Y#-. < = 8D1$T#Y#%=>1%1 1TKTKTK TK .34 TK 5s#\12 TK TKr,rcneZdZdZej Zddfddfddfddfddfddfd dfd dfd dfd d fdd fdd fdd fdd fdZeje jdfdede e e fde eee eefeffddffd Zdedeeeeeee feeeee fffdZdedefdZxZS) AutoAugmentaAutoAugment data augmentation method based on `"AutoAugment: Learning Augmentation Strategies from Data" `_. This transformation works on images and videos only. If the input is :class:`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, optional): Desired policy enum defined by :class:`torchvision.transforms.autoaugment.AutoAugmentPolicy`. Default is ``AutoAugmentPolicy.IMAGENET``. interpolation (InterpolationMode, optional): 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. c0tjdd|SNr^333333?r;linspacenum_binsheightwidths r+zAutoAugment.5>>#sH3Ur,Tc0tjdd|Srrrs r+rzAutoAugment.rr,c6tjdd|z|SNr^gt?rrs r+rzAutoAugment.ENN3 PU@UW_,`r,c6tjdd|z|Srrrs r+rzAutoAugment.ENN3 PV@VX`,ar,c0tjdd|SNr^g>@rrs r+rzAutoAugment.5>>#tX3Vr,c0tjdd|SNr^?rrs r+rzAutoAugment.u~~c3PX7Yr,c0tjdd|Srrrs r+rzAutoAugment.%..c82Tr,c0tjdd|Srrrs r+rzAutoAugment.U^^Ch5Wr,c0tjdd|Srrrs r+rzAutoAugment.ennS#x6Xr,c|dtj||dz dz z z jjSNrr;arangeroundr:rs r+rzAutoAugment.4Q%,,x2HXXY\]^L^2_-`,g,g,i,m,m,or,Fc0tjdd|SNr_r^rrs r+rzAutoAugment.rr,cyr%r7rs r+rzAutoAugment.r,cyr%r7rs r+rzAutoAugment.Tr,cyr%r7rs r+rzAutoAugment.s4r,)r]rerfrgrhrirkrmrorqrsrurvrwNpolicyr!r"r#cbt|||||_|j||_yNr )r&r'r _get_policies _policies)r)rr!r"r*s r+r'zAutoAugment.__init__s0 }4@ ++F3r,c|tjk(rgdS|tjk(rgdS|tjk(rgdSt d|d)N)))rq皙?r)rh333333? )rsrrurNrv皙?NrvrN))rqr)rqrrvrN)rs皙?r)rrhrr))rsrr))rqrrrvr_N))rhrr)rsrr)r)rqrr)rrkrr))rhrrr))rvr^NrrwrNr)rkrr)rmr_r)r)rkr_))rkrr)rsrr))rorrr))r]rrr)rrrrrrr))rw皙?N)rmrr))rhffffff?r)rfrr))rorr)rorr))re?rrgrr))rurNrvrN))rerr)rqrr))rkrr)rirr))rorr)rirr)r)rvrN))rmrr)rorr))rkrr)rfrr))rvrN)rurN))rgrr)rorr))rirr)rkrr))rsrr)rwr^N)rvrNr)rr))rkrrr)rurN)rsrr))rirr)rkrr))rsrrrurN))rgrrr)r)rsrr)rr)rr))r]rr)rwrN)rerrrwrN)r)rsrrrwrNrr)rhrr)rr)r)rwrN))rerr)rsrr)rrr)r)rsrr))rerrr)r)rgrrr))rmrrrhrr)rwrN)rgr^r))rerr)rsrr)rr))rerr)rfrr))r]rrr))rsrr)rgrr))rerrr))r]rr)rgrr))rerr)rurN))r]rrrzThe provided policy z is not recognized.)rIMAGENETCIFAR10SVHNr1)r)rs r+rzAutoAugment._get_policiessk &// / 6(00 0 6(-- - 83F8;NOP Pr,r@c |j|\}}t|\}}|jtt j t |jd}|D]\}}} t jd|ks |j|\} } | d||} | .t| | } | r t jddkr| dz} nd} |j||| |j|j}|j||S)Nr7 rr^r )rTrrr:r;r<r=rand_AUGMENTATION_SPACErrr!r(rW)r)r@rUrSrrrrY probability magnitude_idx magnitudes_fnsigned magnitudesrZs r+forwardzAutoAugment.forwardCs040X0XY_0`-~ 0 EMM#dnn2Er$J KL8> 4L+}::b>[0$($<$<\$J !M6&r659J%!*]";< ejjn3OI !AA itGYGY`d`j`jBN9?$889NP^__r,)r3rr__doc__ _transformsr_v1_transform_clsrrrrrr r:rrr rr'rrrrrrrrrs@r+rrsw$$//VW[\UW[\ `  b  WX\]Y[_`TVZ[WY]^XZ^_ o  XY^_=uE95A7?/8%6$>$>7H7P7PDH 4!4.344ItE$)$4i$?@@A 4  4XQ'XQ eE#uhsm34eCQT `_. This transformation works on images and videos only. If the input is :class:`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, optional): Number of augmentation transformations to apply sequentially. magnitude (int, optional): Magnitude for all the transformations. num_magnitude_bins (int, optional): The number of different magnitude values. interpolation (InterpolationMode, optional): 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. cyr%r7rs r+rzRandAugment.vrr,Fc0tjdd|Srrrs r+rzRandAugment.wrr,Tc0tjdd|Srrrs r+rzRandAugment.xrr,c6tjdd|z|Srrrs r+rzRandAugment.zrr,c6tjdd|z|Srrrs r+rzRandAugment.~rr,c0tjdd|Srrrs r+rzRandAugment.rr,c0tjdd|Srrrs r+rzRandAugment.rr,c0tjdd|Srrrs r+rzRandAugment.rr,c0tjdd|Srrrs r+rzRandAugment.rr,c0tjdd|Srrrs r+rzRandAugment.rr,c|dtj||dz dz z z jjSrrrs r+rzRandAugment.rr,c0tjdd|Srrrs r+rzRandAugment.rr,cyr%r7rs r+rzRandAugment.rr,cyr%r7rs r+rzRandAugment.rr,r\r]rerfrgrhrirkrmrorqrsrurvrrNnum_opsrZnum_magnitude_binsr!r"r#cRt|||||_||_||_yr)r&r'rrZr)r)rrZrr!r"r*s r+r'zRandAugment.__init__s- }4@ ""4r,r@c|j|\}}t|\}}t|jD]}|j |j \}\}} ||j ||} | 8t| |j} | r tjddkr| dz} nd} |j||| |j|j}|j||SNr7rrr^r )rTrrangerr?rrrrZr;rrr!r(rW) r)r@rUrSrr_rYrrrrZs r+rzRandAugment.forwards040X0XY_0`-~ 0 t||$A484I4I$JbJb4c 1L1=&&t'>'>NJ%!*T^^"<= ejjn3OI !AA itGYGY`d`j`jBN%889NP^__r,)r3rrrrrrrrrr:r rrr rr'rrrrs@r+rr^s<*$//95AUW[\UW[\ `  b  WX\]Y[_`TVZ[WY]^XZ^_ o  XY^_=uE95A/8"$7H7P7PDH 5 5 5 5 .34 5 ItE$)$4i$?@@A 5  5`s`s`r,rceZdZdZej Zddfddfddfddfddfd dfd dfd dfd dfd dfddfddfddfddfdZdejdfde de ee fde e e e eefe ffffd ZdedefdZxZS)TrivialAugmentWideaWDataset-independent data-augmentation with TrivialAugment Wide, as described in `"TrivialAugment: Tuning-free Yet State-of-the-Art Data Augmentation" `_. This transformation works on images and videos only. If the input is :class:`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, optional): The number of different magnitude values. interpolation (InterpolationMode, optional): 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. cyr%r7rs r+rzTrivialAugmentWide.rr,Fc0tjdd|SNr^gGz?rrs r+rzTrivialAugmentWide.rr,Tc0tjdd|Sr$rrs r+rzTrivialAugmentWide.rr,c0tjdd|SNr^g@@rrs r+rzTrivialAugmentWide.u~~c4QY7Zr,c0tjdd|Sr'rrs r+rzTrivialAugmentWide.r(r,c0tjdd|S)Nr^g`@rrs r+rzTrivialAugmentWide.s5>>#uh3Wr,c0tjdd|Sr$rrs r+rzTrivialAugmentWide.r(r,c0tjdd|Sr$rrs r+rzTrivialAugmentWide.s%..dH2Ur,c0tjdd|Sr$rrs r+rzTrivialAugmentWide.sU^^Cx5Xr,c0tjdd|Sr$rrs r+rzTrivialAugmentWide.sennS$PX6Yr,c|dtj||dz dz z z jjS)Nrrrrrs r+rzTrivialAugmentWide.rr,c0tjdd|Srrrs r+rzTrivialAugmentWide.rr,cyr%r7rs r+rzTrivialAugmentWide.rr,cyr%r7rs r+rzTrivialAugmentWide.rr,rrNrr!r"c6t|||||_yr)r&r'r)r)rr!r"r*s r+r'zTrivialAugmentWide.__init__s }4@"4r,r@r#c |j|\}}t|\}}|j|j\}\}}||j||} | Ut | t tj|jd} |r tjddkr| dz} nd} |j||| |j|j}|j||Sr)rTrr?rrrr:r;r<rrr!r(rW) r)r@rUrSrrrYrrrrZs r+rzTrivialAugmentWide.forwards040X0XY_0`-~ 0 040E0EdF^F^0_- -}f"4#:#:FEJ  !jU]]4;R;RTV-W)XYZI%**R.C/R I== L)4CUCU\`\f\f> 889NP^__r,)r3rrrrr!rrrrr:r rrr rr'rrrrs@r+r!r!s$$6695AVX\]VX\]Z\`aZ\`aWY]^Z\`aUW[\XZ^_Y[_` o  XY^_=uE95A#,#%7H7P7PDH 55.345ItE$)$4i$?@@A 5`s`s`r,r!ceZdZUdZej Zddfddfddfddfddfdd fd d fd d fd d fd ZieddfddfddfddfdZe e e e e e e geej feffed<dddddej(dfde de de dededeee fdeee eee feffddffd Zd!ej dej fd"Zd#edefd$ZxZS)%AugMixa@AugMix data augmentation method based on `"AugMix: A Simple Data Processing Method to Improve Robustness and Uncertainty" `_. This transformation works on images and videos only. If the input is :class:`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, optional): The severity of base augmentation operators. Default is ``3``. mixture_width (int, optional): The number of augmentation chains. Default is ``3``. chain_depth (int, optional): The depth of augmentation chains. A negative value denotes stochastic depth sampled from the interval [1, 3]. Default is ``-1``. alpha (float, optional): The hyperparameter for the probability distributions. Default is ``1.0``. all_ops (bool, optional): Use all operations (including brightness, contrast, color and sharpness). Default is ``True``. interpolation (InterpolationMode, optional): 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. c0tjdd|Srrrs r+rzAugMix.rr,Tc0tjdd|Srrrs r+rzAugMix.rr,c6tjd|dz |SNr^g@rrs r+rzAugMix.su~~c5SV;X`7ar,c6tjd|dz |Sr:rrs r+rzAugMix.su~~c6TWrr,c|dtj||dz dz z z jjS)Nrrrrs r+rzAugMix.rr,Fc0tjdd|Srrrs r+rzAugMix.rr,cyr%r7rs r+rzAugMix.rr,cyr%r7rs r+rzAugMix.rr,) r]rerfrgrhrqrsrurvc0tjdd|Srrrs r+rzAugMix.rr,c0tjdd|Srrrs r+rzAugMix. rr,c0tjdd|Srrrs r+rzAugMix.!rr,c0tjdd|Srrrs r+rzAugMix."rr,)rirkrmrorrrr_Nseverity mixture_width chain_depthalphaall_opsr!r"r#ct|||d|_d|cxkr|jksntd|jd|d||_||_||_||_||_y)Nr rrz!The severity must be between [1, z]. Got z instead.) r&r'_PARAMETER_MAXr1rErFrGrHrI) r)rErFrGrHrIr!r"r*s r+r'zAugMix.__init__%s} }4@ X4!4!44@ATAT@UU\]e\ffopq q  *&  r,r4c,tj|Sr%)r;_sample_dirichlet)r)r4s r+rMzAugMix._sample_dirichlet9s&&v..r,r@c |j|\}}t|\}}t|tjr|}nt j |}|jr |jn |j}t|j}t|tjrdnd} |jdgt| |j z dz|z} | j#dgdg| j dz zz} |j%tj&|j(|j(g| j*j-| dd} |j%tj&|j(g|j.z| j*j-| dd| dddfj| ddgz} | dddfj| | z}t1|j.D]7}| }|j2dkDr |j2n.t5tj6dddj9}t1|D]}|j;|\}\}}||j<||}|Ut?|t5tj6|j@d }|r tjBd d kr|dz}nd }|jE||||jF|jH }|jK| dd|fj| |z:|j|jM|jN }t|tjPtjfrtjR||}n9t|tTjPjPrt jV|}|jY||S)Nrrrr)devicer)r)lowhighsizer7rr^r )r)like)-rTrr/r;rrx pil_to_tensorrIr_PARTIAL_AUGMENTATION_SPACElistshaperrIreshapemaxndimrRrMtensorrHrOexpandrFrrGr:r<itemr?rKrrErrr!r(add_torrGwraprH to_pil_imagerW)r)r@rUorig_image_or_videorrrSaugmentation_space orig_dims expected_ndimbatch batch_dimsmcombined_weightsmixiaugdepthrrYrrrrZs r+rzAugMix.forward=sd595]5]^d5e22 !45  )5<< 80N__-@AN9=T554KkKk--. '(;Z=M=MNTU &&sSATAT1TVW-X'X[d'dejjm_sejj1n'==  " " LL$**djj1%,, G N NzZ[}^` a   11 LL$**(:(::5<< P W WXbcdXegi j adGOOZ]B/ 011gooj)E1t))*AC(,(8(81(+>N ) %jU]]4==RT5U1V&W XI%**R.C"7!R #I::y@R@RY]YcYc;" HH%ad+33J?#E F#+$kk)$''n.B.B'C )J,<,N>N+O P//#,?@C +SYY__ =..%C889NPSTTr,)r3rrrrr6rrUrrrrrr:rr;rr__annotations__rBILINEARrr rr r'rMrrrrs@r+r6r6s.$**VW[\UW[\acghbdhiVX\] o  XY^_=uE95A #f %fY[_`TVZ[WY]^XZ^_ fc53S/8ELLCY2Y)Z\`)`#aab7H7Q7QDH     .34ItE$)$4i$?@@A (/ //6Us6Us6Ur,r6)3rztypingrrrrrrr r PIL.ImagerHr;torch.utils._pytreer r r torchvisionrrrtorchvision.transformsrrtorchvision.transforms.v2rrrxrr.torchvision.transforms.v2.functional._geometryr*torchvision.transforms.v2.functional._metar+torchvision.transforms.v2.functional._utilsrr_utilsrrrrrrGrIrrrrr!r6r7r,r+rzs JJJ FF=<ffO?OJJU\\399??J4D4DjFVFVVW XKyXKvk`"k`\O`"O`dB`)B`J~U ~Ur,