^jTAHdZddlZddlmZmZmZddlZddlmZddl m Z m Z m Z ddl mZmZmZddlmZmZmZmZmZmZmZmZmZddlmZmZmZdd lm Z d d e e d d fdee!ee!e!ffde"dee#dfdee#dfde$de$f dZ%d dddddddd dddde e ddddd d d d ddd d fdee!ee!e!ffdeee#e#fdeee#e#fd ee"d!e#d"e#d#ee#ee#dffd$ee#d%e$d&e#d'e#d(ee"de"dee#dfdee#dfd)e#d*e"d+e!d,e!de$de$d-e$d.e$d/ee!ee!e!ffd0e!d1e$d2e$f6d3Z&d dddd e e d d d ddd d fdee!ee!e!ffd4ee#d5ee"d6ee!de"dee#dfdee#dfde$de$d.e$d/ee!ee!e!ffd0e!d1e$d2e$fd7Z'd d d ddddddddddd e e dddddddd d d d d ddd d f d8ee!ee!e!fee!e!e!ffd9e$d:e$d ee"deee#e#fdeee#e#fd!e#d"e#d#ee#ee#dffd$ee#d&e#d'e#d(ee"de"dee#dfdee#dfd)e#d*e"d+e!d,e!d4ee#d5ee"d6ee!d;e$de$de$d-e$d.e$d/ee!ee!e!ffd0e!d1e$d2e$f@d<Z(y)=z Transforms Factory Factory methods for building image transforms for use with TIMM (PyTorch Image Models) Hacked together by / Copyright 2019, Ross Wightman N)OptionalTupleUnion) transforms)IMAGENET_DEFAULT_MEANIMAGENET_DEFAULT_STDDEFAULT_CROP_PCT)rand_augment_transformaugment_and_mix_transformauto_augment_transform) str_to_interp_modestr_to_pil_interp!RandomResizedCropAndInterpolationResizeKeepRatioCenterCropOrPadRandomCropOrPad TrimBorder MaybeToTensorMaybePILToTensor)RandomResizedCropToSequenceResizeToSequencePatchify) RandomErasingbilinearFTimg_size interpolationmean.stduse_prefetcher normalizec||dk(rd}tj|t|tj|g}|r|t gz }n[|s|t gz }nJ|t tj tj|tj|gz }tj|S)a No-augmentation image transforms for training. Args: img_size: Target image size. interpolation: Image interpolation mode. mean: Image normalization mean. std: Image normalization standard deviation. use_prefetcher: Prefetcher enabled. Do not convert image to tensor or normalize. normalize: Normalization tensor output w/ provided mean/std (if prefetcher not used). Returns: randomrrrr) rResizer CenterCroprr NormalizetorchtensorCompose)rrrrr r!tfls g/var/www/ramen.bs-engineer-server.com/venv/lib/python3.12/site-packages/timm/data/transforms_factory.pytransforms_noaug_trainr.s* " (2D]2STh' C  "##   "## O  \\$'LL%      c ""?g?r#consti@scaleratiotrain_crop_modehflipvflip color_jittercolor_jitter_probforce_color_jittergrayscale_probgaussian_blur_prob auto_augmentre_probre_modere_count re_num_splitsseparatenaflex patch_size max_seq_lenpatchifypatchify_channels_lastc|xsd}|dvsJg}|r3t|xsd}t|xsd}|t||||| gz }n|dvrSt|xsd}t|xsd}|t|| d |d d | |d k(r t|d n t |d gz }n1t|xsd}t|xsd}|t |||| gz }|dkDr|t j|gz }|dkDr|t j|gz }g}d}| rt| tsJ|xsd| v }t|ttfr t|}n|}tt|dzt| Dcgc]}tdtd|zc}} | r| dk7rt!| | d<| j#dr|t%| | gz }n7| j#drd| d<|t'| | gz }n|t)| | gz }|{|syt|ttfrt+|dvsJt-|fdz}|-|t j.t j0|g|gz }n|t j0|gz }| r|t j2| gz }| r0|t j.t j4d g| gz }g}!|r|!t7gz }!nt|s|!t7gz }!nc|!t9t j:t=j>| t=j>|!gz }!|dkDr|!tA||||d"#gz }!|r|!tC||$gz }!|r>t jD|t jD|t jD|!fSt jD||z|!zScc}w)%a ImageNet-oriented image transforms for training. Args: img_size: Target image size. train_crop_mode: Training random crop mode ('rrc', 'rkrc', 'rkrr'). 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). force_color_jitter: Force color jitter where it is normally disabled (ie with RandAugment on). 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). interpolation: Image interpolation mode. mean: Image normalization mean. std: Image normalization standard deviation. re_prob: Random erasing probability. re_mode: Random erasing fill mode. re_count: Number of random erasing regions. re_num_splits: Control split of random erasing across batch size. use_prefetcher: Prefetcher enabled. Do not convert image to tensor or normalize. normalize: Normalize tensor output w/ provided mean/std (if prefetcher not used). separate: Output transforms in 3-stage tuple. naflex: Enable NaFlex mode, sequence constrained patch output patch_size: Patch size for NaFlex mode. max_seq_len: Max sequence length for NaFlex mode. Returns: If separate==True, the transforms are returned as a tuple of 3 separate transforms for use in a mixing dataset that passes * all data through the first (primary) transform, called the 'clean' data * a portion of the data through the secondary transform * normalizes and converts the branches above with the third, final transform rrc>rKrkrcrkrr)g{Gz??)g?gUUUUUU?)rFrGr5r6r)rLrM)g?rN)g?grq?r0T)rrandom_scale_probrandom_scale_rangerandom_scale_arearandom_aspect_probrandom_aspect_rangerLreflect) padding_mode)r5r6rr1)pF3ag?)translate_constimg_meanr#rrandaugmixg333333? translate_pct)r^) kernel_sizer%cpu)mode max_count num_splitsdevicerF channels_last)#tuplerrrrrrRandomHorizontalFlipRandomVerticalFlip isinstancestrlistmindictintroundr startswithr r r lenfloat RandomApply ColorJitterRandomGrayscale GaussianBlurrrr(r)r*rrr+)"rr5r6r7r8r9r:r;r<r=r>r?rrrr@rArBrCr r!rDrErFrGrHrI primary_tfl secondary_tfldisable_color_jitter img_size_minx aa_params final_tfls" r-transforms_imagenet_trainrAs[D&.O 5 55 5K e*{+e1123!#'     . .%.;/E%.;/E "/&)',&*'*(-#f, yAyA  K%.;/E%5#56E 1"/  K rz 77%@AA  rz 55>?? M ,,,,%7$N$,:NO h .x=L#L t 34dCCU37^4CD  ]h6):=)IIo &  " "6 * 4\9MN NM  $ $X .),Io & 7 iPQ QM 4\9MN NM(< lT5M 2|$. ..",/1A5L  ( &&".. =(( M j44lCD DM*44~FGG   " "++;$%   I&()) &())  O  \\$'LL%    R<  &,  Ih*DZ[\\ !!+. 0B0B=0QS]SeSefoSppp!!+ "= "IJJYDs*Ocrop_pct crop_modecrop_border_pixelsc 6|xst}t|ttfr@t |dk(sJt|Dcgc]}t j ||z c}}nt j ||z }||f}g}|r|t|gz }| r|t| | |gz }n|dk(r9|tj|t|tj|gz }n|dk(r:|Dcgc]}td|z}}|t||dt|| gz }n[|d |d k(r(|tj|d t|gz }n|t|gz }|tj|gz }|r|t!gz }n[|s|t!gz }nJ|t#tj$t'j(|t'j(| gz }| r|t+| | gz }tj,|Scc}wcc}w)a ImageNet-oriented image transform for evaluation and inference. Args: img_size: Target image size. crop_pct: Crop percentage. Defaults to 0.875 when None. crop_mode: Crop mode. One of ['squash', 'border', 'center']. Defaults to 'center' when None. crop_border_pixels: Trim a border of specified # pixels around edge of original image. interpolation: Image interpolation mode. mean: Image normalization mean. std: Image normalization standard deviation. use_prefetcher: Prefetcher enabled. Do not convert image to tensor or normalize. normalize: Normalize tensor output w/ provided mean/std (if prefetcher not used). naflex: Enable NaFlex mode, sequence constrained patch output patch_size: Patch size for NaFlex mode. max_seq_len: Max sequence length for NaFlex mode. patchify: Patchify the output instead of relying on prefetcher Returns: Composed transform pipeline )rFrGrsquashr$borderrXrN)rlongest)fillrr3r%rg)r rlrirnrtmathfloorrrrr&r r'rrrrrrr(r)r*rr+)rrrrrrrr r!rErFrGrHrIr~ scale_sizer,vrs r-transforms_imagenet_evalrs:H++H(UDM*8}!!!hGDJJq8|4GH ZZ8 34  *- C -.//   !#'      !!*TUVV   c ""H62s H,H input_size is_trainingno_augtf_preprocessingc t|ttfr|dd} n|} |r|r|rJdddlm}!|!|| | }"|"S|r|r|rJdt | | ||||}"|"S|rWt | fid |d |d |d |d |d|d| d| d| d| d| d|d|d|d|d|d|d|d|d|d|d|d|d |d!|}"|"S|rJd"t| | ||||||||||||#}"|"S)$a; Args: input_size: Target input size (channels, height, width) tuple or size scalar. is_training: Return training (random) transforms. no_aug: Disable augmentation for training (useful for debug). train_crop_mode: Training random crop mode ('rrc', 'rkrc', 'rkrr'). 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). interpolation: Image interpolation mode. mean: Image normalization mean. std: Image normalization standard deviation. re_prob: Random erasing probability. re_mode: Random erasing fill mode. re_count: Number of random erasing regions. re_num_splits: Control split of random erasing across batch size. 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. tf_preprocessing: Use TF 1.0 inference preprocessing for testing model ports use_prefetcher: Pre-fetcher enabled. Do not convert image to tensor or normalize. normalize: Normalization tensor output w/ provided mean/std (if prefetcher not used). separate: Output transforms in 3-stage tuple. Returns: Composed transforms or tuple thereof Nz6Separate transforms not supported for TF preprocessingr)TfPreprocessTransform)rsizerz-Cannot perform split augmentation with no_aug)rrrr r!r7r5r6r8r9r:r;r=r>r?rrrr@rArBrCr r!rDrErFrGrHrIz>Separate transforms not supported for validation preprocessing) rrrrrrr r!rErFrGrHrI)rlrirntimm.data.tf_preprocessingrr.rr)#rrrr7r5r6r8r9r:r;r=r>r?rrrr@rArBrCrrrrr r!rDrErFrGrHrIrr transforms# r-create_transformr{sJ*udm,bc?NUUU|D)#' @ u 6 P!P P<.+-# Ip a1 /     *#4 .$6*,   !""#$,%& .'($)*"+,-.&/0(12"34(>5I^ % a!a a<0+!##5-#%'!'=I" r/))__doc__rtypingrrrr) torchvisionrtimm.data.constantsrrr timm.data.auto_augmentr r r timm.data.transformsr rrrrrrrrtimm.data.naflex_transformsrrrtimm.data.random_erasingrrqrmruboolr.rrrr/r-rs"  )) "]]llccc__214'"7!5$ *#U38_,-*#*#E3J*#5#:  *#  *#  *#\14/3/3)-8;-1#( "$&&*%"7!5$24'+7MKU38_,-MKeUl+,MKeUl+,MK"# MK  MK  MKE5#445MK$E?MK!MKMK"MKsmMKMKE3JMK5#: MK !MK"#MK$%MK&'MK()MK*+MK,-MK./MK0#uS#X./1MK23MK45MK6!%7MKb14$(#',0'"7!5$24'+g#U38_,-g#5/g#C=g#%SM g#  g# E3J g#5#: g#g#g#g##uS#X./g#g#g#!%g#VIL!)-/3/38;-1 "$&&*'"7!5$(#',0!&$24'+AM#uS#Xc3m0DDEMMM"# M eUl+, M eUl+, MMME5#445M$E?MM"MsmMME3JM 5#: !M"#M$%M&'M()M*5/+M,C=-M.%SM/M01M23M45M67M89M:#uS#X./;M<=M>?M@!%AMr/