RЦi?ONSrSSKrSSKrSSKJr SSKJr SSKJr SSK J r J r J r J r SSKrSSKrSSKrSSKJrJr SS KJrJr SS KJrJr SS KJr SS KJr SS KJ r \RB"\"5r#Sr$Sr%"SS5r&SISjr'SSSSSSSSSSSSSSSSSSS\\SSSSSSSS\RP\RR"S5SSSSS4$S\ \\4S\ \*\ \*\*4\ \*\*\*44S\*S\+S \+S!\,S"\-S#\*S$\+S%\ \-S&\ \ \,\,4S'\ \ \,\,4S(\,S)\,S*\,S+\ \,S,\,S-\,S.\ \-S/\*S0\*S1\-S2\ \,S34S4\ \,S34S5\*S6\+S7\ \,S8\ \-S9\ \*S:\ \ S;\+S<\+S=\R\S>\RRS?\+S@\+SA\+SB\-SC\+4NSDjjr/"SESF\R`RbRd5r3"SGSH\45r5g)JaLoader Factory, Fast Collate, CUDA Prefetcher Prefetcher and Fast Collate inspired by NVIDIA APEX example at https://github.com/NVIDIA/apex/commit/d5e2bb4bdeedd27b1dfaf5bb2b24d6c000dee9be#diff-cf86c282ff7fba81fad27a559379d5bf Hacked together by / Copyright 2019, Ross Wightman N)suppress)partial)repeat)CallableOptionalTupleUnion)IMAGENET_DEFAULT_MEANIMAGENET_DEFAULT_STD)IterableImageDataset ImageDataset)OrderedDistributedSamplerRepeatAugSampler) RandomErasing)FastCollateMixup)create_transformc N[US[5(de[U5n[USS[5(Ga,[USS[R5n[USS5nX-n[ R "U[ RS9n[ R "U/USSSRQ7[ RS9n[U5Hn[XS5U:Xde[U5H`nXSXWX--'U(a/XgX--==[ R"XSU5- ss'MGXgX--==XSU- ss'Mb M Xe4$[USS[R5(a[ R"UV s/sHoSPM sn [ RS9n[U5U:Xde[ R "U/USSRQ7[ RS9n[U5H(nXg==[ R"XS5- ss'M* Xe4$[USS[ R5(a[ R"UV s/sHoSPM sn [ RS9n[U5U:Xde[ R "U/USSRQ7[ RS9n[U5HnXgRXS5 M Xe4$es sn fs sn f)zcA fast collation function optimized for uint8 images (np array or torch) and int64 targets (labels)r)dtyper ) isinstancetuplelennpndarraytorchzerosint64shapeuint8range from_numpytensorTensorcopy_) batch batch_sizeis_npinner_tuple_sizeflattened_batch_sizetargetsr"ijbs O/var/www/html/ai-image-ml/venv/lib/python3.13/site-packages/timm/data/loader.py fast_collater/s eAh & && &UJ%(1+u%%58A; 3uQx{+)<++2%++F2JU1Xa[^5I5IJRWR]R]^z"Aux{#'77 77+,.3hqkAN*+q~-.%2B2B58A;q>2RR.q~-.%(1+a.@. -# E!HQK , ,,,e4e!e4EKKH7|z)))j=58A;+<+<=U[[Qz"A I))%(1+6 6I# E!HQK . .,,e4e!e4EKKH7|z)))j=58A;+<+<=U[[Qz"A IOOEHQK (#u 5 5s 'LL"c<[U[[45(d[[X55nU$[ U5U:waE[ R "U5R5nU4U-n[RSUS35 U$[ U5U:XdS5eU$)Nz@Pretrained mean/std different shape than model, using avg value .z-normalization stats must match image channels) rrlistrrrmeanitem_loggerwarning)xnx_means r. adapt_to_chsr:Ds a% ' ' &,  H Q1" IMZ[\Z]]^_` H1v{KKK{ Hc`\rSrSr\\S\R"S5SSSSSS 4 S \RRRS \ \ S 4S \ \ S 4S\ S\RS\\RS\S\ S\S\ S\ 4SjjrSrSr\S5r\S5r\S5r\R2S5rSrg)PrefetchLoaderPcudaNFconstr rloaderr3.stdchannelsdevice img_dtypefp16re_probre_modere_count re_num_splitsc [X$5n[X45nSUSS4n XlXPlU(a[RnU=(d [R Ul[R"UV s/sHoS-PM sn XVS9RU 5Ul [R"UV s/sHoS-PM sn XVS9RU 5Ul US:a[UU U U US9Ul OSUl URS:H=(a [RR5UlURS:H=(a [R"R5Ulgs sn fs sn f)Nr )rFrrA) probabilitymode max_count num_splitsrFr@npu)r:rCrFrfloat16float32rGr"viewr3rDrrandom_erasingtyper@ is_availableis_cudarSis_npu)selfrCr3rDrErFrGrHrIrJrKrLnormalization_shaper7s r.__init__PrefetchLoader.__init__Rs0D+3) (Aq1   I"3emmLL" #dWd #FEEITJ]E^ <<! "cWc "6DDHDI\D]  R<"/#"( #D #'D {{f,J1H1H1J kkU*Guyy/E/E/G  $ "s 2E).E.c## SnUR(aJ[RRURS9n[ [RR US9nOcUR(aJ[RRURS9n[ [RR US9nOSn[nURGH[upEU"5 URURSS9nURURSS9nURUR5RUR5RUR 5nUR"bUR#U5nSSS5 U(dWW4v OSnUbUR(a7[RR%URS9R'U5 OGUR(a6[RR%URS9R'U5 UnUnGM^ WW4v g!,(df  N=f7f)NT)rF)stream)rF non_blockingF)rZrr@StreamrFrrar[rSrrCtorGsub_r3div_rDrWcurrent_stream wait_stream)r\firstrastream_context next_input next_targetinputtargets r.__iter__PrefetchLoader.__iter__{s <<ZZ&&dkk&:F$UZZ%6%6vFN [[YY%%T[[%9F$UYY%5%5fENF%N'+{{ #J!']]$++D]Q )nnDKKdnS ']]4>>:?? JOOPTPXPXY &&2!%!4!4Z!@J "Vm#!<<JJ--T[[-AMMfU[[II,,DKK,@LLVTE F+(3.Vm+"!s CIB H6 ? ?;;))77 7r;c[URR[5(aXRRlggrrr~)r\r7s r.rrs0 dkk,,.> ? ?34KK " " 0 @r;)rFrGrZr[rCr3rWrD)__name__ __module__ __qualname____firstlineno__r r rrFutilsdata DataLoaderrfloatintrrboolstrr^rortpropertyrwr{rsetter__static_attributes__r;r.r=r=Ps9 '<%9#(<<#7/3"!"'HKK$$//'Hs #'Hucz" 'H  'H LL 'H  ,'H'H'H'H'H'HR#J #### 55r;r=allc[RRR5nURU:Xde[ U[ 5(aWU"U5n[R"U5 [R"U5 [RRUS-5 gUS;deUS:Xa-[RRURS-5 gg)Nl)rpartr) rrrget_worker_infoidrrrandomseed manual_seedr) worker_idworker_seeding worker_infors r. _worker_initrs++""224K >>Y && &.(++k* D $ t{+,000 U " IINN;++{; < #r;FrArBg?g?bilinearr@Tr{ input_sizer& is_trainingno_augrIrJrKre_splittrain_crop_modescaleratiohflipvflip color_jittercolor_jitter_probgrayscale_probgaussian_blur_prob auto_augmentnum_aug_repeatsnum_aug_splits interpolationr3.rD num_workers distributedcrop_pct crop_modecrop_border_pixelsr pin_memoryrHrGrFuse_prefetcheruse_multi_epochs_loaderpersistent_workersrtf_preprocessingc'Sn'U(a U=(d Sn'[U40SU_SU_SU _SU _SU _SU _S U _S U_S U_S U_S U_SU_SU_SU_SU_SU_SU_SU_SU_SU_SU_SU'_SU&_SU"_SUS:_6Ul[U[5(aUR US9 Sn(U(a[U[ R RR5(dYU(aFU(a [UUS9n(OM[ R RRRU5n(O[U5n(O US:XdS5eUc;U"(a[O-[ R RRRn[ R RR n)U#(a["n)[%U[U[ R RR5(+=(a U(SL=(a UUU(UUU['[(U%S 9U$S!9 n*U)"U40U*D6n+U"(a(U(a U(dUOS#n-[/U+UUUSU!UU U-UUU'S$9 n+U+$![*a$n,U*R-S"5 U)"U40U*D6n+Sn,A,NZSn,A,ff=f)%a Args: dataset: The image dataset to load. input_size: Target input size (channels, height, width) tuple or size scalar. batch_size: Number of samples in a batch. is_training: Return training (random) transforms. no_aug: Disable augmentation for training (useful for debug). re_prob: Random erasing probability. re_mode: Random erasing fill mode. re_count: Number of random erasing regions. re_split: Control split of random erasing across batch size. scale: Random resize scale range (crop area, < 1.0 => zoom in). ratio: Random aspect ratio range (crop ratio for RRC, ratio adjustment factor for RKR). hflip: Horizontal flip probability. vflip: Vertical flip probability. color_jitter: Random color jitter component factors (brightness, contrast, saturation, hue). Scalar is applied as (scalar,) * 3 (no hue). color_jitter_prob: Apply color jitter with this probability if not None (for SimlCLR-like aug grayscale_prob: Probability of converting image to grayscale (for SimCLR-like aug). gaussian_blur_prob: Probability of applying gaussian blur (for SimCLR-like aug). auto_augment: Auto augment configuration string (see auto_augment.py). num_aug_repeats: Enable special sampler to repeat same augmentation across distributed GPUs. num_aug_splits: Enable mode where augmentations can be split across the batch. interpolation: Image interpolation mode. mean: Image normalization mean. std: Image normalization standard deviation. num_workers: Num worker processes per DataLoader. distributed: Enable dataloading for distributed training. crop_pct: Inference crop percentage (output size / resize size). crop_mode: Inference crop mode. One of ['squash', 'border', 'center']. Defaults to 'center' when None. crop_border_pixels: Inference crop border of specified # pixels around edge of original image. collate_fn: Override default collate_fn. pin_memory: Pin memory for device transfer. fp16: Deprecated argument for half-precision input dtype. Use img_dtype. img_dtype: Data type for input image. device: Device to transfer inputs and targets to. use_prefetcher: Use efficient pre-fetcher to load samples onto device. use_multi_epochs_loader: persistent_workers: Enable persistent worker processes. worker_seeding: Control worker random seeding at init. tf_preprocessing: Use TF 1.0 inference preprocessing for testing model ports. Returns: DataLoader rrrrrrrrrrrrrrr3rDrrrrIrJrKrLrrseparate)rN) num_repeatszORepeatAugment not currently supported in non-distributed or IterableDataset use)r) r&shufflerrwrr drop_lastworker_init_fnrrrA) r3rDrErFrHrGrIrJrKrL)r transformrr set_loader_cfgrrrIterableDatasetrrDistributedSamplerrr/ dataloaderdefault_collaterMultiEpochsDataLoaderdictrr TypeErrorpopr=).r{rr&rrrIrJrKrrrrrrrrrrrrrrr3rDrrrrrrrrHrGrFrrrrrrLrw loader_class loader_argsrCeprefetch_re_probs. r. create_loaderrsnM&+! ((     ",&."$ !"#$%&.'()*+,-.$/0*12&34 !#5G:'/00 ;7G:gu{{/?/?/O/OPP *7P++**66II'R08G!#v%vv#%3\9I9I9T9T9d9d ;;##..L, w (8(8(H(HIImgY]ombm|NK- K6g55&1&7b ]$'   M' 6,-g556s I I2I--I2c4^\rSrSrU4SjrSrSrSrU=r$)ri}c>[TU]"U0UD6 SUlURc[ UR 5UlO[ UR5UlSUl[TU]5Ulg)NFT)superr^_DataLoader__initialized batch_sampler_RepeatSamplerrwroiterator)r\argskwargs __class__s r.r^MultiEpochsDataLoader.__init__sc $)&)(-%    %)$,,7DL!/0B0B!CD (,%(* r;cURc[UR5$[URR5$rr)rrrwrss r.rtMultiEpochsDataLoader.__len__s2$($6$6$>s4<< cCHZHZHbHbDccr;c#p# [[U55Hn[UR5v M g7frr)r rnextr)r\r+s r.roMultiEpochsDataLoader.__iter__s's4y!At}}% %"s46)rrrrw) rrrrr^rtror __classcell__)rs@r.rr}s+d&&r;rc$\rSrSrSrSrSrSrg)riz;Sampler that repeats forever. Args: sampler (Sampler) cXlgrrrw)r\rws r.r^_RepeatSampler.__init__s r;c#N# [UR5ShvN MN7frr)iterrwrss r.ro_RepeatSampler.__iter__s"DLL) ) ) )s %#%rN)rrrr__doc__r^rorrr;r.rrs *r;r)r)6rloggingr contextlibr functoolsr itertoolsrtypingrrrr rtorch.utils.datanumpyr constantsr r r{r rdistributed_samplerrrrWrmixuprtransforms_factoryr getLoggerrr5r/r:r=rrUrFrrrrrrrrrrobjectrrr;r.rsJ 33 B7L)#0   H %$N  e5e5P =(")-/3/3!-1 "$&&* '"7!5!$(#',0)- !&$||F3#(-#'#!&Or|%99:r#uS#Xc3m0DDErr r  r  rrrr"#reUl+,reUl+,rrrr $E?!r"#r$"%r&sm'r()r*+r,-r.E3J/r05#: 1r23r45r65/7r8C=9r:%SM;r<X&=r>?r@ArB;;CrD ErFGrH"&IrJ!KrLMrNOrj&EKK,,77&( *V *r;