U ¯Ê¹jÞŠã@stdZddlZddlZddlZddlmZddlmZmZm Z m Z ddl m Z m Z mZmZmZddl Z ddlZedd„e j d¡dd …DƒƒZd Zd Zed ed Zee dƒrÆe jje jjfZe jjZ ne je jfZe jZ dd„Z!dd„Z"dd„Z#dd„Z$dd„Z%dd„Z&dd„Z'dd„Z(dd „Z)d!d"„Z*d#d$„Z+d%d&„Z,d'd(„Z-d¨d*d+„Z.d,d-„Z/d.d/„Z0d0d1„Z1d2d3„Z2d4d5„Z3d6d7„Z4d8d9„Z5d:d;„Z6dd?„Z8d@dA„Z9dBdC„Z:d©dGdH„Z;dIdJ„ZdOdP„Z?dQdR„Z@dSdT„ZAdUdV„ZBdWdX„ZCdYdZ„ZDddde8e?e@eAeBeCeDe9e:e9e:e9e:e9e:ee>ee;d[dEd\ee;d]d^d\e;d_œZEe*e,e+e)e/e/e/e-e-e.e1e1e0e0e2e2e3e3e#e$e'e(e%e&e6e4e5d_œZFGd`da„daƒZGdbdc„ZHddde„ZIdfdg„ZJdhdi„ZKdjdk„ZLdªdmdn„ZMGdodp„dpƒZNd«eOe edqœdrds„ZPdtdudvdwdxdydzd{d|d}d~dd€dd‚gZQdtdudvdwdƒd„dzd…d†d‡dˆdd€dd‚gZRd„d‰dŠgZSd‹d‹d‹dŒd d dddddddddŽœZTdŒd d ddddddddddddd‘œZUed’œd“d”„ZVd¬eOd•œd–d—„ZWd­e eXeYfeYe ee e eefd™œdšd›„ZZGdœd„dƒZ[d®eOe ee e eOeefdžœdŸd „Z\dtd…d†d‡dˆdudwdƒd„dd€dd‚g Z]d¯e eXeYfe ee e eOeefd¡œd¢d£„Z^Gd¤d¥„d¥ƒZ_d°eOe edqœd¦d§„Z`dS)±a AutoAugment, RandAugment, AugMix, and 3-Augment for PyTorch This code implements the searched ImageNet policies with various tweaks and improvements and does not include any of the search code. AA and RA Implementation adapted from: https://github.com/tensorflow/tpu/blob/master/models/official/efficientnet/autoaugment.py AugMix adapted from: https://github.com/google-research/augmix 3-Augment based on: https://github.com/facebookresearch/deit/blob/main/README_revenge.md Papers: AutoAugment: Learning Augmentation Policies from Data - https://arxiv.org/abs/1805.09501 Learning Data Augmentation Strategies for Object Detection - https://arxiv.org/abs/1906.11172 RandAugment: Practical automated data augmentation... - https://arxiv.org/abs/1909.13719 AugMix: A Simple Data Processing Method to Improve Robustness and Uncertainty - https://arxiv.org/abs/1912.02781 3-Augment: DeiT III: Revenge of the ViT - https://arxiv.org/abs/2204.07118 Hacked together by / Copyright 2019, Ross Wightman éN)Úpartial)ÚDictÚListÚOptionalÚUnion)ÚImageÚImageOpsÚ ImageEnhanceÚ ImageChopsÚ ImageFiltercCsg|] }t|ƒ‘qS©)Úint)Ú.0Úxr r úU/var/www/html/bwave/python/venv/lib/python3.8/site-packages/timm/data/auto_augment.pyÚ "srÚ.é)é€rrg$@éú)Útranslate_constÚimg_meanÚ ResamplingcCs(| dt¡}t|ttfƒr$t |¡S|S)NÚresample)ÚpopÚ_DEFAULT_INTERPOLATIONÚ isinstanceÚlistÚtupleÚrandomÚchoice)ÚkwargsÚ interpolationr r rÚ_interpolation5s  r#cCs*d|krtdkr| d¡t|ƒ|d<dS)NÚ fillcolor©érr)Ú_PIL_VERrr#)r!r r rÚ_check_args_tf<s r(c Ks*t|ƒ|j|jtjd|ddddff|ŽS©Nér©r(Ú transformÚsizerÚAFFINE©ÚimgÚfactorr!r r rÚshear_xBsr2c Ks*t|ƒ|j|jtjddd|ddff|ŽSr)r+r/r r rÚshear_yGsr3c Ks8||jd}t|ƒ|j|jtjdd|dddff|ŽS)Nrr*©r-r(r,rr.©r0Zpctr!Úpixelsr r rÚtranslate_x_relLsr7c Ks8||jd}t|ƒ|j|jtjddddd|ff|ŽSr)r4r5r r rÚtranslate_y_relRsr8c Ks*t|ƒ|j|jtjdd|dddff|ŽSr)r+©r0r6r!r r rÚtranslate_x_absXsr:c Ks*t|ƒ|j|jtjddddd|ff|ŽSr)r+r9r r rÚtranslate_y_abs]sr;c Ks t|ƒtdkr|j|f|ŽStdkr|j\}}d}|d|df}t |¡ }tt |¡dƒtt |¡dƒdtt |¡ dƒtt |¡dƒdg}dd„} | |d  |d |d  |d |ƒ\|d <|d <|d |d 7<|d |d 7<|j |jt j |f|ŽS|j||d dS)N)r&rr%)rrç@éçc Ss8|\}}}}}}||||||||||fS©Nr ) rÚyÚmatrixÚaÚbÚcÚdÚeÚfr r rr,tszrotate..transformrr*rr&r)r) r(r'Úrotater-ÚmathÚradiansÚroundÚcosÚsinr,rr.) r0Údegreesr!ÚwÚhZ post_transZ rotn_centerZanglerAr,r r rrHbs2   ú ÿrHcKs t |¡Sr?)rZ autocontrast©r0Ú__r r rÚ auto_contrastsrScKs t |¡Sr?)rÚinvertrQr r rrT…srTcKs t |¡Sr?)rÚequalizerQr r rrU‰srUcKs t ||¡Sr?)rÚsolarize)r0ÚthreshrRr r rrVsrVrcKstg}tdƒD],}||kr.| td||ƒ¡q | |¡q |jdkrp|jdkrft|ƒdkrf|||}| |¡S|S)Nééÿ)ÚLÚRGBr[)ÚrangeÚappendÚminÚmodeÚlenÚpoint)r0ÚaddrWrRZlutÚir r rÚ solarize_add‘s     rdcKs|dkr |St ||¡S)Né)rÚ posterize)r0Z bits_to_keeprRr r rrf¡srfcKst |¡ |¡Sr?)r ÚContrastÚenhance©r0r1rRr r rÚcontrast§srjcKst |¡ |¡Sr?)r ÚColorrhrir r rÚcolor«srlcKst |¡ |¡Sr?)r Ú Brightnessrhrir r rÚ brightness¯srncKst |¡ |¡Sr?)r Ú Sharpnessrhrir r rÚ sharpness³srpcKs| tj|d¡}|S)N©Zradius)Úfilterr Ú GaussianBlurrir r rÚ gaussian_blur·srtcKs*d}d}| tjt |||¡d¡}|S)Nçš™™™™™¹?r<rq)rrr rsrÚuniform)r0r1rRZ radius_minZ radius_maxr r rÚgaussian_blur_rand¼srwcKs$tdtdd|ƒƒ}t |¡ |¡S)Nçð?r>)r^Úmaxr rkrh)r0r1Ú_r r rÚ desaturateÃsr{cCst ¡dkr| S|S)zWith 50% prob, negate the valueçà?)r)Úvr r rÚ_randomly_negateÉsr~cCs|td}t|ƒ}|fS)Ng>@©Ú _LEVEL_DENOMr~©ÚlevelÚ_hparamsr r rÚ_rotate_level_to_argÎs r„cCs|tddfS)NgÍÌÌÌÌÌü?ru)r€rr r rÚ_enhance_level_to_argÕsr…cCs$|td}tddt|ƒƒ}|fS)NgÍÌÌÌÌÌì?rurx)r€ryr~rr r rÚ _enhance_increasing_level_to_argÚs r†r>rxTcCs2|t}||||}|r,t|t||ƒƒ}|fSr?)r€ryr^)r‚rƒÚmin_valÚmax_valÚclampr r rÚ_minmax_level_to_argâs rŠcCs|td}t|ƒ}|fS)Ng333333Ó?rrr r rÚ_shear_level_to_argês r‹cCs&|d}|tt|ƒ}t|ƒ}|fS)Nr)r€Úfloatr~)r‚Úhparamsrr r rÚ_translate_abs_level_to_argñsrŽcCs&| dd¡}|t|}t|ƒ}|fS)NÚ translate_pctgÍÌÌÌÌÌÜ?)Úgetr€r~)r‚rrr r rÚ_translate_rel_level_to_argøs  r‘cCst|tdƒfS©Né©r r€rr r rÚ_posterize_level_to_argsr•cCsdt||ƒdfS)Nr“r)r•)r‚rr r rÚ"_posterize_increasing_level_to_argsr–cCst|tdƒdfSr’r”rr r rÚ _posterize_original_level_to_argsr—cCstdt|tdƒƒfS)NrX©r^r r€rr r rÚ_solarize_level_to_argsr™cCsdt||ƒdfS)NrXr)r™rr r rÚ!_solarize_increasing_level_to_argsršcCstdt|tdƒƒfS)Nrénr˜rr r rÚ_solarize_add_level_to_arg!srœr|)r‡rˆrur<)Ú AutoContrastÚEqualizeÚInvertÚRotateÚ PosterizeÚPosterizeIncreasingÚPosterizeOriginalÚSolarizeÚSolarizeIncreasingÚ SolarizeAddrkÚColorIncreasingrgÚContrastIncreasingrmÚBrightnessIncreasingroÚSharpnessIncreasingÚShearXÚShearYZ TranslateXZ TranslateYÚ TranslateXRelÚ TranslateYRelÚ DesaturatersÚGaussianBlurRandc@s&eZdZd dd„Zdd„Zdd „ZdS) Ú AugmentOpr|é NcCsŠ|pt}||_t||_t||_||_||_| ¡|_ t d|krJ|dnt d|kr\|dnt d|_ |j  dd¡|_|j  dd¡|_dS)Nrr")r$rÚ magnitude_stdrÚ magnitude_max)Ú_HPARAMS_DEFAULTÚnameÚ NAME_TO_OPÚaug_fnÚ LEVEL_TO_ARGÚlevel_fnÚprobÚ magnitudeÚcopyrÚdictÚ_FILLÚ_RANDOM_INTERPOLATIONr!rr³r´)Úselfr¶r»r¼rr r rÚ__init__gs   þ zAugmentOp.__init__cCs¬|jdkrt ¡|jkr|S|j}|jdkr`|jtdƒkrHt d|¡}n|jdkr`t ||j¡}|jpht}t dt ||ƒƒ}|j dk r’|  ||j ¡nt ƒ}|j|f|ž|jŽS)NrxrÚinfr>)r»rr¼r³rŒrvÚgaussr´r€ryr^rºrrr¸r!)rÁr0r¼Ú upper_boundZ level_argsr r rÚ__call__|s   zAugmentOp.__call__cCsZ|jjd|j›d|j›}|d|j›d|j›7}|jdk rN|d|j›7}|d7}|S)Nz(name=z, p=z, m=z, mstd=z, mmax=ú))Ú __class__Ú__name__r¶r»r¼r³r´)rÁÚfsr r rÚ__repr__Žs  zAugmentOp.__repr__)r|r²N©rÉÚ __module__Ú __qualname__rÂrÆrËr r r rr±es r±cs°ddgddgddgddgd d gd d gd dgddgddgddgddgddgddgddgddgddgd d!gd"d#gd$d%gd&d'gd(d)gd*d+gd,d-gdd.gd/d0gg}‡fd1d2„|Dƒ}|S)3N©ržçš™™™™™é?r*©r¬rÐr“©rkçš™™™™™Ù?é ©ržç333333ã?é©rkrÓr*©r rÖre©r¤rÐrשržrÓé©r¤rÓr©r¤rÖr©rkçš™™™™™É?r©ržrÐre©ržrÓre©r¦rÐrשr«ràrÔ©rkrÖr*©ržrxr©rŸrÓrÔ©r rÖr©ržrxrÔ©r¬rÖrשrkrÓrÜ©ržrÖr)r¡rÓé©rrÓrÜ©r¤rÖre©rkrÖrÔ©r¤ràr“©r rÐrÔ©r rxrÜ©r®rÐrÔ©r«r>r©r¤rÐr“©r¬rÐr©rkrÖr“©rkrxr©r rÖr©ržrÐr“©ržr>re©ržrxr“©rrÖr©r¬rÓrÜ©r¦rÖrÜ)r¡rÐr©r¤rÖr²©ržrÖr*©rkrÐrí©r rÓr&csg|]}‡fdd„|Dƒ‘qS)csg|]}t|dˆiŽ‘qS©r©r±©rrBrr rr´sz5auto_augment_policy_v0...r ©rÚsprr rr´sz*auto_augment_policy_v0..r ©rÚpolicyZpcr rrÚauto_augment_policy_v0—s8çr cs°ddgddgddgddgd d gd d gd dgddgddgddgddgddgddgddgddgddgd d!gd"d#gd$d%gd&d'gd(d)gd*d+gd,d-gdd.gd/d0gg}‡fd1d2„|Dƒ}|S)3NrÏrÑrÒrÕrØrÙrÚrÛrÝrÞrßrárârãrärårærçrèrérêrërì©r¢rÓrírîrïrðrñròrórôrõrör÷rørùrúrûrürýrþrÿr)r¢rÐrrrrrcsg|]}‡fdd„|Dƒ‘qS)csg|]}t|dˆiŽ‘qSrrrrr rrÖsz6auto_augment_policy_v0r...r rrr rrÖsz+auto_augment_policy_v0r..r r r rrÚauto_augment_policy_v0r¸s8çrcs°ddgddgddgddgd d gd d gd dgddgddgddgd dgddgddgddgddgd dgddgd d!gd"d#gddgd d gddgddgddgddgg}‡fd$d%„|Dƒ}|S)&N)r£rÓre©r rÖrÔ©r¤rÖr&©rrÖr&rárÕ)r£rÖrÜ)r£rÖrírÛrñ©ržrÓr“©r rÐre©r¤rÖrשržrÖrÜ)r£rÐr&ræ©r ràr×rï©ržrÖre)r£rÓrí©rkrÓr©r rÓrÔ©ržrÖr©ržr>rÜ©rŸrÖr“©ržrxrerø©rgrxre©rkrxr©rkrÐre©r¤rÐrÜ©rorÓrÜ©rŸrÖre©r«rÖr&récsg|]}‡fdd„|Dƒ‘qS)csg|]}t|dˆiŽ‘qSrrrrr rr÷sz;auto_augment_policy_original...r rrr rr÷sz0auto_augment_policy_original..r r r rrÚauto_augment_policy_originalÚs8çr%cs°ddgddgddgddgd d gd d gd dgddgddgddgd dgddgddgddgddgd dgddgd d!gd"d#gddgd d gddgddgddgddgg}‡fd$d%„|Dƒ}|S)&N)r¢rÓrerrrrárÕ)r¢rÖrÜ)r¢rÖrírÛrñrrrr)r¢rÐr&rærrïrr rrrrrrrørrr r!r"r#r$récsg|]}‡fdd„|Dƒ‘qS)csg|]}t|dˆiŽ‘qSrrrrr rrsz..r rrr rrsz1auto_augment_policy_originalr..r r r rrÚauto_augment_policy_originalrûs8çr&cs&dgdgdgg}‡fdd„|Dƒ}|S)N)r¤rxr&)r¯rxr²)r°rxr²csg|]}‡fdd„|Dƒ‘qS)csg|]}t|dˆiŽ‘qSrrrrr rr"sz5auto_augment_policy_3a...r rrr rr"sz*auto_augment_policy_3a..r r r rrÚauto_augment_policy_3as ýr'Úv0cCsn|pt}|dkrt|ƒS|dkr(t|ƒS|dkr8t|ƒS|dkrHt|ƒS|dkrXt|ƒSdsjtd|›ƒ‚dS)NÚoriginalZ originalrr(Zv0rÚ3aFzUnknown AA policy )rµr%r&r rr'ÚAssertionError)r¶rr r rÚauto_augment_policy&sr,c@s$eZdZdd„Zdd„Zdd„ZdS)Ú AutoAugmentcCs ||_dSr?)r )rÁr r r rrÂ7szAutoAugment.__init__cCs"t |j¡}|D] }||ƒ}q|Sr?)rr r )rÁr0Z sub_policyÚopr r rrÆ:s  zAutoAugment.__call__cCsL|jjd}|jD],}|d7}|d dd„|Dƒ¡7}|d7}q|d7}|S)Nz(policy=z [z, cSsg|] }t|ƒ‘qSr )Ústr)rr.r r rrDsz(AutoAugment.__repr__..ú]rÇ)rÈrÉr Újoin)rÁrÊÚpr r rrË@s   zAutoAugment.__repr__NrÌr r r rr-5sr-)Ú config_strrc CsŒ| d¡}|d}|dd…}|D]T}t d|¡}t|ƒdkr@q"|dd…\}}|dkrj| dt|ƒ¡q"d s"td ƒ‚q"t||d }t|ƒS) az Create a AutoAugment transform Args: config_str: String defining configuration of auto augmentation. Consists of multiple sections separated by dashes ('-'). The first section defines the AutoAugment policy (one of 'v0', 'v0r', 'original', 'originalr'). The remaining sections: 'mstd' - float std deviation of magnitude noise applied Ex 'original-mstd0.5' results in AutoAugment with original policy, magnitude_std 0.5 hparams: Other hparams (kwargs) for the AutoAugmentation scheme Returns: A PyTorch compatible Transform ú-rr*Nú(\d.*)rÚmstdr³Fz"Unknown AutoAugment config sectionr)ÚsplitÚrer`Ú setdefaultrŒr+r,r-) r3rÚconfigZ policy_namerDÚcsÚkeyÚvalZ aa_policyr r rÚauto_augment_transformJs     r>rržrŸr r¡r¤r¦rkrgrmror«r¬r­r®r¢r¥r§r¨r©rªr¯rsrír×r*)r¥r¯rsr r«r¬r¢rr§rªr¨r©ržrŸgÐ?gš™™™™™©?)r r«r¬r­r®r§rªrr¥r¦r¨r©ržr¢rŸ)Ú transformscCs4tt| ¡Žƒ\}}t |¡}|t |¡}||fSr?)rÚzipÚitemsÚnpÚarrayÚsum)r?Zprobsr r rÚ_get_weighted_transformsÃs rE)r¶cCs0|dkr tS|dkrtS|dkr$tS|r,tStS)NÚweightsZ3awr*)Ú_RAND_WEIGHTED_0Ú_RAND_WEIGHTED_3AÚ_RAND_3AÚ_RAND_INCREASING_TRANSFORMSÚ_RAND_TRANSFORMS)r¶Ú increasingr r rÚrand_augment_choicesÊsrMr²©r¼r»rr?cs&ˆpt‰|pt}‡‡‡fdd„|DƒS)Ncsg|]}t|ˆˆˆd‘qS)©r»r¼rr©rr¶©rr¼r»r rrÜsÿÿz$rand_augment_ops..)rµrKrNr rQrÚrand_augment_opsÔs ÿrRc@s&eZdZd dd„Zdd„Zdd„ZdS) Ú RandAugmentrNcCs||_||_||_dSr?)ÚopsÚ num_layersÚchoice_weights)rÁrTrUrVr r rrÂászRandAugment.__init__cCs6tjj|j|j|jdk|jd}|D] }||ƒ}q$|S)N)Úreplacer2)rBrr rTrUrV)rÁr0rTr.r r rrÆæsü zRandAugment.__call__cCs<|jjd|j›d}|jD]}|d|›7}q|d7}|S)Nz(n=ú, ops=ú rÇ)rÈrÉrUrT©rÁrÊr.r r rrËòs  zRandAugment.__repr__)rNrÌr r r rrSàs  rS)r3rr?cCs¤t}d}d}d}| d¡}|ddks*t‚|dd…}|D]ü}| d ¡rft|dd…ƒ} |dkrd| }q:t d |¡} t| ƒdkr€q:| dd…\} } | d kr¾t| ƒ} | d kr°td ƒ} | d| ¡q:| dkrØ| dt | ƒ¡q:| dkrît | ƒrìd}q:| dkrt | ƒ}q:| dkrt | ƒ}q:| dkr*t| ƒ}q:ds:tdƒ‚q:t |tƒrRt ||d}n|dkrj|rft nt}d} t |tƒr†t|ƒ\}} t||||d}t||| dS)a+ Create a RandAugment transform Args: config_str (str): String defining configuration of random augmentation. Consists of multiple sections separated by dashes ('-'). The first section defines the specific variant of rand augment (currently only 'rand'). The remaining sections, not order sepecific determine 'm' - integer magnitude of rand augment 'n' - integer num layers (number of transform ops selected per image) 'p' - float probability of applying each layer (default 0.5) 'mstd' - float std deviation of magnitude noise applied, or uniform sampling if infinity (or > 100) 'mmax' - set upper bound for magnitude to something other than default of _LEVEL_DENOM (10) 'inc' - integer (bool), use augmentations that increase in severity with magnitude (default: 0) 't' - str name of transform set to use Ex 'rand-m9-n3-mstd0.5' results in RandAugment with magnitude 9, num_layers 3, magnitude_std 0.5 'rand-mstd1-tweights' results in mag std 1.0, weighted transforms, default mag of 10 and num_layers 2 hparams (dict): Other hparams (kwargs) for the RandAugmentation scheme Returns: A PyTorch compatible Transform rFr|r4rZrandr*NÚtr5r6édrÃr³Zmmaxr´ÚincTÚmÚnr2z"Unknown RandAugment config section)rLrN)rV)r€r7r+Ú startswithr/r8r`rŒr9r ÚboolrrMrJrKrrErRrS)r3rr?r¼rUrLr»r:rDr=r;r<r6rVZra_opsr r rÚrand_augment_transformúsT               rb©r¼rr?cs$ˆpt‰|pt}‡‡fdd„|DƒS)Ncsg|]}t|dˆˆd‘qS)rxrOrrP©rr¼r rrcsûüzaugmix_ops..)rµÚ_AUGMIX_TRANSFORMSrcr rdrÚ augmix_ops\s  ûrfc@sBeZdZdZddd„Zdd „Zd d „Zd d „Zdd„Zdd„Z dS)Ú AugMixAugmenta AugMix Transform Adapted and improved from impl here: https://github.com/google-research/augmix/blob/master/imagenet.py From paper: 'AugMix: A Simple Data Processing Method to Improve Robustness and Uncertainty - https://arxiv.org/abs/1912.02781 rxr×éÿÿÿÿFcCs"||_||_||_||_||_dSr?)rTÚalphaÚwidthÚdepthÚblended)rÁrTrirjrkrlr r rrÂqs zAugMixAugment.__init__cCs\||}d}g}|ddd…D]"}||}|d|9}| |¡qtj|ddd…tjdS)Nrxrhr*©Zdtype)r]rBrCÚfloat32)rÁÚwsr^ZcumpZrwsrOrir r rÚ_calc_blended_weightsxs  z#AugMixAugment._calc_blended_weightsc Csx| ¡}| ||¡}|D]Z}|jdkr,|jn tj dd¡}tjj|j|dd}|} |D] } | | ƒ} qVt  || |¡}q|S)Nrr*r“T©rW) r½rprkrBrÚrandintr rTrÚblend) rÁr0Úmixing_weightsr^Zimg_origrorOrkrTÚimg_augr.r r rÚ_apply_blended‚s  zAugMixAugment._apply_blendedc CsÊ|jd|jdt| ¡ƒf}tj|tjd}|D]d}|jdkrF|jn tj dd¡}tjj |j |dd}|} |D] } | | ƒ} qp||tj | tjd7}q2tj |dd|dt  | tj¡¡}t  |||¡S) Nrr*rmr“Trqgào@)Úout)r-r`ZgetbandsrBZzerosrnrkrrrr rTZasarrayZcliprZ fromarrayZastypeZuint8rs) rÁr0rtr^Z img_shapeÚmixedÚmwrkrTrur.r r rÚ _apply_basic’s zAugMixAugment._apply_basiccCs\t tj |jg|j¡¡}t tj |j|j¡¡}|jrJ| |||¡}n|  |||¡}|Sr?) rBrnrZ dirichletrirjÚbetarlrvrz)rÁr0rtr^rxr r rrÆ£s zAugMixAugment.__call__cCsL|jjd|j›d|j›d|j›d}|jD]}|d|›7}q,|d7}|S)Nz(alpha=z, width=z, depth=rXrYrÇ)rÈrÉrirjrkrTrZr r rrˬs & zAugMixAugment.__repr__N)rxr×rhF) rÉrÍrÎÚ__doc__rÂrprvrzrÆrËr r r rrgks   rgc Csd}d}d}d}d}| d¡}|ddks.t‚|dd …}|D]®}t d |¡} t| ƒd kr\q>| d d …\} } | d kr†| d t| ƒ¡q>| dkr˜t| ƒ}q>| dkrªt| ƒ}q>| dkr¼t| ƒ}q>| dkrÎt| ƒ}q>| dkràt| ƒ}q>ds>tdƒ‚q>| d tdƒ¡t||d} t | ||||dS)a  Create AugMix PyTorch transform Args: config_str (str): String defining configuration of random augmentation. Consists of multiple sections separated by dashes ('-'). The first section defines the specific variant of rand augment (currently only 'rand'). The remaining sections, not order sepecific determine 'm' - integer magnitude (severity) of augmentation mix (default: 3) 'w' - integer width of augmentation chain (default: 3) 'd' - integer depth of augmentation chain (-1 is random [1, 3], default: -1) 'b' - integer (bool), blend each branch of chain into end result without a final blend, less CPU (default: 0) 'mstd' - float std deviation of magnitude noise applied (default: 0) Ex 'augmix-m5-w4-d2' results in AugMix with severity 5, chain width 4, chain depth 2 hparams: Other hparams (kwargs) for the Augmentation transforms Returns: A PyTorch compatible Transform r×rhrxFr4rZaugmixr*Nr5rr6r³r^rOrErBrCzUnknown AugMix config sectionrÃ)r¼r)rirjrkrl) r7r+r8r`r9rŒr rarfrg) r3rr¼rjrkrirlr:rDr;r<r=rTr r rÚaugment_and_mix_transform´s:          r})r)r>rxT)r(N)N)T)r²r|NN)NN)r²NN)N)ar|rrIr8Ú functoolsrÚtypingrrrrZPILrrr r r ÚnumpyrBrÚ __version__r7r'r¿r€r¾rµÚhasattrrZBILINEARZBICUBICrÀrr#r(r2r3r7r8r:r;rHrSrTrUrVrdrfrjrlrnrprtrwr{r~r„r…r†rŠr‹rŽr‘r•r–r—r™ršrœr¹r·r±r rr%r&r'r,r-r/r>rKrJrIrHrGrErMr rŒrRrSrbrerfrgr}r r r rÚsÐ "þ       ä!å2!"!! $ññýòñ ü ü ýý Róý ý I