RЦiSrSSKrSSKrSSKrSSKJr SSKJrJrJ r J r SSK J r J r JrJrJr SSK r SSKr\"\ R(R+S5SSVs/sH n\"U5PM sn5rSrS r\"S \S 9r\"\ S 5(aC\ R:R<\ R:R>4r \ R:R>r!O$\ R<\ R>4r \ R>r!S r"Sr#Sr$Sr%Sr&Sr'Sr(Sr)Sr*Sr+Sr,Sr-Sr.SnSjr/Sr0Sr1Sr2Sr3Sr4S r5S!r6S"r7S#r8S$r9S%r:S&r;SoS(jrS+r?S,r@S-rAS.rBS/rCS0rDS1rE0S2S_S3S_S4S_S5\9_S6\@_S7\A_S8\B_S9\C_S:\D_S;\E_S<\:_S=\;_S>\:_S?\;_S@\:_SA\;_SB\:_\;\=\=\>\>\?\?\"\\1_S?\1_S@\3_SA\3_SB\4_\4\$\%\(\)\&\'\7\5\6SG. ErG"SHSI5rHSJrISKrJSLrKSMrLSNrMSpSOjrN"SPSQ5rOSqSR\PSS\ \4STjjrQ/SUQrR/SVQrS/SWQrTSXSXSXSYSSSZSZSZSZSZSZSZSZS[.rUSYSSSZSZS\S\S\S]S]S]S]S]S]S]S^.rVS_\4S`jrWSrSa\P4SbjjrXSsSc\ \\Y4Sd\YSS\ \S_\ \ \\44SejjrZ"SfSg5r[StSR\PSS\ \S_\ \ \P\\44Shjjr\/SiQr]SuSc\ \\Y4SS\ \S_\ \ \P\\44Sjjjr^"SkSl5r_SqSR\PSS\ \4Smjjr`gs snf)vaAutoAugment, 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 N)partial)DictListOptionalUnion)ImageImageOps ImageEnhance ImageChops ImageFilter.)rrg$@)translate_constimg_mean ResamplingcURS[5n[U[[45(a[ R "U5$U$)Nresample)pop_DEFAULT_INTERPOLATION isinstancelisttuplerandomchoice)kwargs interpolations U/var/www/html/ai-image-ml/venv/lib/python3.13/site-packages/timm/data/auto_augment.py_interpolationr 5s9JJz+ABM-$//}}]++ cbSU;a:aURS5 [U5US'g)N fillcolorrr)_PIL_VERrr )rs r_check_args_tfr'<s,fF!2 ;'/F:r!c ~[U5 UR"UR[RSUSSSS440UD6$Nrr' transformsizerAFFINEimgfactorrs rshear_xr2Bs76 ==5<s rtranslate_y_absrA]s76 ==5<Uup4pVpxX0-XA--U-X`-Xq--U-4$N) xymatrixabcdefs rr,rotate..transformts3!' Q1515=1$aeaema&77 7r!rr*rr%r)r) r'r&rotater-mathradiansroundcossinr,rr.) r0degreesrwh post_trans rotn_centeranglerKr,s rrSrSbsX66zz',V,,6xx 3wC( g&& $((5/2 & $((5/2 &  488E?"B ' $((5/2 &    8 ) ^Ojm +k!n_z!}-Lf q 6!9 q [^# q [^# }}SXXu||VFvFF ::gz(:: ;;r!c .[R"U5$rG)r autocontrastr0__s r auto_contrastrcs   %%r!c .[R"U5$rG)r invertras rreres ??3 r!c .[R"U5$rG)r equalizeras rrgrgs   S !!r!c .[R"X5$rG)r solarize)r0threshrbs rriris   S ))r!c &/n[S5H8nXR:aUR[SXQ-55 M'URU5 M: URS;a7URS:Xa[ U5S:XaXD-U-nUR U5$U$)N)LRGBro)rangeappendminmodelenpoint)r0addrjrblutis r solarize_addrys| C 3Z : JJs3( ) JJqM   xx< 88u SS)c/Cyy~ Jr!c >US:aU$[R"X5$)N)r posterize)r0 bits_to_keeprbs rr|r|s q   c 00r!c L[R"U5RU5$rG)r Contrastenhancer0r1rbs rcontrastrs   % - -f 55r!c L[R"U5RU5$rG)r Colorrrs rcolorrs   c " * *6 22r!c L[R"U5RU5$rG)r Brightnessrrs r brightnessrs  " "3 ' / / 77r!c L[R"U5RU5$rG)r Sharpnessrrs r sharpnessrs  ! !# & . .v 66r!c LUR[R"US95nU$)Nradius)filterr GaussianBlurrs r gaussian_blurrs **[--V< =C Jr!c SnSnUR[R"[R"X4U-5S95nU$)N皙?rCr)rr rruniform)r0r1rb radius_min radius_maxs rgaussian_blur_randrs9JJ **[--V^^J]cPc5de fC Jr!c ~[S[SSU- 55n[R"U5R U5$)N?rE)rrmaxr rr)r0r1_s r desaturaters5 SR&[) *F   c " * *6 22r!c>[R"5S:aU*$U$)zWith 50% prob, negate the value?)r)vs r_randomly_negaters3&A2-A-r!c6U[- S-n[U5nU4$)Ng>@ _LEVEL_DENOMrlevel_hparamss r_rotate_level_to_argr" \ !S (E U #E 6Mr!c"U[- S-S-4$)Ng?r)rrs r_enhance_level_to_argrs L C '# - ..r!cPU[- S-n[SS[U5-5nU4$)Ng?rr)rrrrs r _enhance_increasing_level_to_argrs1\ !R 'E S+E22 3E 6Mr!rcfU[- nX#U- U--nU(a[U[X055nU4$rG)rrrr)rrmin_valmax_valclamps r_minmax_level_to_argrs9 \ !E (E1 1E GS01 6Mr!c6U[- S-n[U5nU4$)Ng333333?rrs r_shear_level_to_argrrr!cRUSnU[- [U5-n[U5nU4$)Nr)rfloatr)rhparamsrs r_translate_abs_level_to_argrs2/0O \ !U?%; ;E U #E 6Mr!cZURSS5nU[- U-n[U5nU4$)N translate_pctg?)getrr)rrrs r_translate_rel_level_to_argrs2KK6M \ !] 2E U #E 6Mr!c.[U[- S-54$Nintrrs r_posterize_level_to_argrs  $) * ++r!c&S[X5S- 4$)Nrr)r)rrs r"_posterize_increasing_level_to_argrs &u6q9 9 ::r!c4[U[- S-5S-4$rrrs r _posterize_original_level_to_argrs!  $) *Q . //r!cB[S[U[- S-554$)Nrlrrrrrs r_solarize_level_to_argrs# sC-45 6 77r!c&S[X5S- 4$)Nrlr)rrs r!_solarize_increasing_level_to_argrs '8; ; <,>+?@@    ) GD../0 0B c  r!) rrrrrrrrr)r Nr __module__ __qualname____firstlineno__rrr__static_attributes__rHr!rrresE*<$r!rcSS/SS/SS/SS/S S /S S /S S/SS/SS/SS/SS/SS/SS/SS/SS/SS/S S!/S"S#/S$S%/S&S'/S(S)/S*S+/S,S-/SS./S/S0//nUVVs/sHo"Vs/sHn[US1U06PM snPM nnnU$s snfs snnf)2Nr皙?r*rrrr皙? r333333?rrr*rrr{rrrrrrrrrrrr皙?rrrr{rrr{rrrrrrrrr*rrrrrrrrrrrrrrrrrrrrr)rrrrrrrr{rrrrrrrrrrrrrrrrrErrrrrrrrrrrrrrrrrrrrrEr{rrrrrrrrrrrr)rrrrrrrrr*rrr+rrr%rrrpolicysprLpcs rauto_augment_policy_v0rHs* 12 01 ./ 34 34 01 67 /0 01 /0 12 01  89 01 12 67 12 ./ ./ 34 78 45  56 34 ./3F6BH H22 62a9a ) )2 6B H I 7 H BA= 1B=BcSS/SS/SS/SS/S S /S S /S S/SS/SS/SS/SS/SS/SS/SS/SS/SS/S S!/S"S#/S$S%/S&S'/S(S)/S*S+/S,S-/SS./S/S0//nUVVs/sHo"Vs/sHn[US1U06PM snPM nnnU$s snfs snnf)2Nr rrrrrrrrrrrr r!r"r#r$r%r&r'r(r)r*rrr+r,r-r.r/r0r1r2r3r4r5r6r7r8r9r:r;r<r=r>)rrrr?r@rArBrrCrDs rauto_augment_policy_v0rrLs* 12 01 ./ 34 34 01 67 /0 01 /0 12 01 (*BC 01 12 67 12 ./ ./ 34 78 45 (*?@ 34 ./3F6BH H22 62a9a ) )2 6B H I 7 HrIcSS/SS/SS/SS/S S /S S /S S/SS/SS/SS/S S/SS/SS/SS/SS/S S/SS/S S!/S"S#/SS/S S /SS/SS/SS/SS//nUVVs/sHo"Vs/sHn[US$U06PM snPM nnnU$s snfs snnf)%N)rrr{rrrrrr%rrr%rr)rrr)rrr+rr/rrrrrr{rrrrrr)rrr%r$rrrr-rrr{)rrr+rrrrrrrrrrrErrrrrrr{r6rrr{rrrrrr{rrrrrrrrr{rrr%r'rrCrDs rauto_augment_policy_originalrds* '(:; 78 34 &(EF 34 12 34 &(<= 12 <= ./ 12 34 12 01 ./ 01  23 12 01 34 78 12 01 343F6BH H22 62a9a ) )2 6B H I 7 HrIcSS/SS/SS/SS/S S /S S /S S/SS/SS/SS/S S/SS/SS/SS/SS/S S/SS/S S!/S"S#/SS/S S /SS/SS/SS/SS//nUVVs/sHo"Vs/sHn[US$U06PM snPM nnnU$s snfs snnf)%N)rrr{rNrOrPrr)rrr)rrr+rr/rQrRrSrT)rrr%r$rUr-rVrKrWrXrYrZr[r\r6r]r^r_r`rarbrcr'rrCrDs rauto_augment_policy_originalrrfs* )*<= 78 34 (*IJ 34 12 34 (*>? 12 >? ./ 12 34 12 01 ./ 01  23 12 01 34 78 12 01 343F6BH H22 62a9a ) )2 6B H I 7 HrIc S/S/S//nUVVs/sHo"Vs/sHn[USU06PM snPM nnnU$s snfs snnf)N)rrr%)rrr)rrrrrCrDs rauto_augment_policy_3arhsY  ! &'F BH H22 62a9a ) )2 6B H I 7 Hs =8 ==cU=(d [nUS:Xa [U5$US:Xa [U5$US:Xa [U5$US:Xa [ U5$US:Xa [ U5$SU35e)Noriginal originalrv0v0r3azUnknown AA policy )rrdrfrHrLrh)rrs rauto_augment_policyro&s{))G z+G44 {,W55 t|%g.. u}&w// t|%g..-&tf--5r!c&\rSrSrSrSrSrSrg) AutoAugmenti5cXlgrGrE)rrEs rrAutoAugment.__init__7s r!ch[R"UR5nUH nU"U5nM U$rG)rrrE)rr0 sub_policyops rrAutoAugment.__call__:s,]]4;;/ BS'C r!c URRS-nURH;nUS- nUSRUVs/sHn[ U5PM sn5- nUS- nM= US- nU$s snf)Nz(policy=z [z, ]r)rrrEjoinstr)rrprws rrAutoAugment.__repr__@sq ^^ $ $z 1A 'MB $))q1qSWq12 2B #IB c  2sA, rsNrrHr!rrqrq5s r!rq config_strrc URS5nUSnUSSnUHZn[R"SU5n[U5S:aM+USSupgUS:XaURS[ U55 MUS 5e [ X1S 9n[ U5$) akCreate 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'). While the remaining sections define other arguments * 'mstd' - float std deviation of magnitude noise applied hparams: Other hparams (kwargs) for the AutoAugmentation scheme Returns: A PyTorch compatible Transform Examples:: 'original-mstd0.5' results in AutoAugment with original policy, magnitude_std 0.5 -rr*N(\d.*)rmstdrz"Unknown AutoAugment config section)r)splitrert setdefaultrrorq) rrconfig policy_namerNcskeyval aa_policys rauto_augment_transformrJs$  c "F)K ABZF  XXi # r7Q; bq6 &=   c ; >> >5$KAI y !!r!)rrrrrrrrrrrrrrr)rrrrrrrrrrrrrrr)rrrr+rr*)rrrrrrrrrrrrrrg?g?)rrrrrrrrrrrrrrr transformsc[[UR565up[R"U5nU[R "U5- nX4$rG)rzipitemsnparraysum)rprobss r_get_weighted_transformsrsES*"2"2"456J HHUOE BFF5M !E  r!rcpUS:Xa[$US:Xa[$US:Xa[$U(a[$[$)Nweights3awrn)_RAND_WEIGHTED_0_RAND_WEIGHTED_3A_RAND_3A_RAND_INCREASING_TRANSFORMS_RAND_TRANSFORMS)r increasings rrand_augment_choicesrs8 y u}   t|*4 &J:JJr!rrc ~U=(d [nU=(d [nUVs/sH n[XAXS9PM sn$s snf)Nrrr)rrr)rrrrrs rrand_augment_opsrsM ))G//JKU WKU4  9 ?KU WW Ws:c*\rSrSrSSjrSrSrSrg) RandAugmentiNc(XlX lX0lgrG)ops num_layerschoice_weights)rrrrs rrRandAugment.__init__s$,r!c[RRURURUR SLUR S9nUH nU"U5nM U$)N)replacer})rrrrrr)rr0rrws rrRandAugment.__call__sZii HH OO''4/!!   BS'C r!cURRSURS3-nURH nUSU3- nM US- nU$)Nz(n=, ops= r)rrrrrrrws rrRandAugment.__repr__sP ^^ $ $T__,=V'D D((B D+ B c  r!)rrr)rNrrHr!rrrs- r!rcj[nSnSnSnURS5nUSS:XdeUSSnUGHnURS 5(a[USS5n UcU nM/M1[R"S U5n [ U 5S:aMYU SSupU S :Xa0[ U 5n U S :a [ S 5n URSU 5 MU S:XaURS[U 55 MU S:Xa[U 5(aSnMMU S:Xa [U 5nMU S:Xa [U 5nMU S:Xa[ U 5nGMS5e [U[5(a [X%S9nOUcU(a[O[nSn [U[5(a [U5up-[!X6XS9n[#XU S9$)aCreate 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 specific 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 hparams (dict): Other hparams (kwargs) for the RandAugmentation scheme Returns: A PyTorch compatible Transform Examples:: '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 rFrrrrandr*NtrrdrrmmaxrincTmnr}z"Unknown RandAugment config section)r)rrrr)r)rr startswithr|rrtrrrboolrrrrrrrr)rrrrrrrrrNrrrrrra_opss rrand_augment_transformrs<IJJ D   c "F !9   ABZF  <<  ae*C! ")Q'B2w{"1vHCf}Sz#: 0DT N*d##%=j%I"  g eF v. IIr!) rrrrrrrrrrrrrc U=(d [nU=(d [nUVs/sHn[USUUS9PM sn$s snf)Nrr)r_AUGMIX_TRANSFORMSr)rrrrs r augmix_opsr_sV ))G11J    $        s<c@\rSrSrSrS SjrSrSrSrSr Sr S r g ) AugMixAugmentinzAugMix 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 c@XlX lX0lX@lXPlgrG)ralphawidthdepthblended)rrrrrrs rrAugMixAugment.__init__ts    r!cX-nSn/nUSSS2H nXS- nUSU- -nURU5 M" [R"USSS2[RS9$)Nrr*dtype)rqrrfloat32)rwsrcumprwsrZrs r_calc_blended_weights#AugMixAugment._calc_blended_weights{se VDbDAHE QY D JJu xxDbD 44r!cxUR5nURX#5nUHnURS:a URO[RR SS5n[RR URUSS9nUn UH n U "U 5n M [R"XU5nM U$)Nrr*rTr) rrrrrrandintrrrblend) rr0mixing_weightsrimg_origrrZrrimg_augrws r_apply_blendedAugMixAugment._apply_blendeds 88:  ' ' :A"&**q.DJJbii6G6G16ME))""488UD"ACGW+++cA.C  r!cURSURS[UR554n[R"U[R S9nUHnUR S:a UR O[RRSS5n[RRURUSS9nUn UH n U "U 5n M XV[R"U [R S9-- nM [R"USSUS9 [R"UR[R 55n[R""XU5$) Nrr*rrTrgo@)out)r-rtgetbandsrzerosrrrrrrasarrayclipr fromarrayastypeuint8r) rr0rr img_shapemixedmwrrrrws r _apply_basicAugMixAugment._apply_basicsHHQK!c#,,..AA "**5 B"&**q.DJJbii6G6G16ME))""488UD"ACGW+ "**WBJJ?? ?E ! q$E* RXX 67{{3q))r!c[R"[RRUR/UR -55n[R"[RR URUR55nUR(aURXU5nU$URXU5nU$rG) rrr dirichletrrbetarrr)rr0rrrs rrAugMixAugment.__call__sBII$7$7 tzz8Q$RS JJryy~~djj$**= > <<''Q?E %%c1=E r!cURRSURSURSURS3-nUR H nUSU3- nM US- nU$)Nz(alpha=z, width=z, depth=rrr)rrrrrrrs rrAugMixAugment.__repr__si ^^ $ $ HTZZLPXY]YcYcXddj'k k((B D+ B c  r!)rrrrrN)rrrF) rrr r __doc__rrrrrrr rHr!rrrns% 5 *"r!rc4SnSnSnSnSnURS5nUSS:XdeUSS nUHn[R"S U5n [U 5S :aM+U S S upU S :XaURS [ U 55 MUU S:Xa [ U 5nMhU S:Xa [ U 5nM{U S:Xa [ U 5nMU S:Xa [ U 5nMU S:Xa [ U 5nMS5e URS [ S55 [X!S9n [XX4US9$)aCreate AugMix PyTorch transform Args: config_str (str): String defining configuration of random augmentation. Consists of multiple sections separated by dashes ('-'). The first section defines the specific variant of rand augment (currently only 'rand'). The remaining sections, not order specific determine 'm' - integer magnitude (severity) of augmentation mix (default: 3) 'w' - integer width of augmentation chain (default: 3) 'd' - integer depth of augmentation chain (-1 is random [1, 3], default: -1) 'b' - integer (bool), blend each branch of chain into end result without a final blend, less CPU (default: 0) 'mstd' - float std deviation of magnitude noise applied (default: 0) Ex 'augmix-m5-w4-d2' results in AugMix with severity 5, chain width 4, chain depth 2 hparams: Other hparams (kwargs) for the Augmentation transforms Returns: A PyTorch compatible Transform rrrFrraugmixr*NrrrrrrZrOrLrMzUnknown AugMix config sectionr)rr)rrrr) rrrtrrrrrr) rrrrrrrrrNrrrrs raugment_and_mix_transformrs$&I E E EG   c "F !9  ABZF  XXi # r7Q; bq6 &=   c ; CZCI CZHE CZHE CZ#JE CZ3iG 99 95'( e 5 y :C W UUr!)r)rErT)rlNrG)T)rrNN)NN)rNN)arrrTr functoolsrtypingrrrrPILrr r r r numpyrr __version__rrr&rrrrhasattrrBILINEARBICUBICrrr r'r2r4r:r<r?rArSrcrergriryr|rrrrrrrrrrrrrrrrrrrrrrrrrHrLrdrfrhrorqr|rrrrrrrrrrrrrrrr)rIs0rrs, ..FF  #//"7"7"& "*  1 6387 3 . / ,;08 = 8 D d "  ( =9&;- "7%: '!"<#$&%&<!!--00.SI0#sK,9 @M f f    9 (< Uu(* J!"#$%!!$$!*7 >//dBDBB .* "s "Xd^ "F(( *&KsK(*"&26 We$ W W$ WU4:./ W8#'7;QJQJ$QJU3d?34QJh$(*"&7; e$ $ U3d?34 FFR1V#1V1VkBsK-