o پiS?@sdZddlZddlmZmZmZddlZddlmZddl 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 ddZ%d dddddddd dddde e ddddd d d d ddd fdee!ee!e!ffdeee#e#fd eee#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$d/e$d0ee!ee!e!ffd1e!d2e$f4d3d4Z&d dddd e e d d d ddd f dee!ee!e!ffd5ee#d6ee"d7ee!de"dee#dfdee#dfde$de$d/e$d0ee!ee!e!ffd1e!d2e$fd8d9Z'd d d ddddddddddd e e dddddddd d d d d ddd fd:ee!ee!e!fee!e!e!ffd;e$dd>d?Z(dS)@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 normalizecCsz|dkrd}tj|t|dt|g}|r|tg7}n|s&|tg7}n|ttjt|t|dg7}t |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)rrrrrr tflr+P/home/ubuntu/.local/lib/python3.10/site-packages/timm/data/transforms_factory.pytransforms_noaug_trains" r-?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_lenpatchifyc CsL|pd}|dvs Jg}|r't|pd}t|pd}|t||||| dg7}nE|dvrUt|p/d}t|p5d}|t|| d |d d |d |d krLt|d dnt|d dg7}nt|pYd}t|p_d}|t|||| dg7}|dkry|tj|dg7}|dkr|tj|dg7}g}d}| rt | t sJ|pd| v }t |tt frt |}n|}t t|dtdd| Dd}| r| dkrt| |d<| dr|t| |g7}n| drd|d<|t| |g7}n|t| |g7}|dur*|s*t |t tfrt|dvsJnt|fd }|dur"|tjtj|g|dg7}n|tj|g7}| r6|tj| dg7}| rH|tjtjd!d"g| dg7}g}|rT|tg7}n-|s^|tg7}n#|ttjt| t|d#g7}|dkr|t||||d$d%g7}|r|t |d&g7}|rt!|t!|t!|fSt!|||S)'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>rGrkrcrkrr)g{Gz??)g?gUUUUUU?)rDrEr3r4r)rHrI)g?rJ)g?grq?r.T)rrandom_scale_probrandom_scale_rangerandom_scale_arearandom_aspect_probrandom_aspect_rangerHreflect) padding_mode)r3r4rr/)pF3ag?cSsg|] }tdtd|qS)minround.0xr+r+r, sz-transforms_imagenet_train..)translate_constimg_meanr!rrandaugmixg333333? translate_pctN)ra) kernel_sizer#cpu)mode max_count num_splitsdevicerD)"tuplerrrrrrRandomHorizontalFlipRandomVerticalFlip isinstancestrlistrVdictintr startswithr r r lenfloat RandomApply ColorJitterRandomGrayscale GaussianBlurrrr&r'r(rrr)) rr3r4r5r6r7r8r9r:r;r<r=rrrr>r?r@rArr rBrCrDrErF primary_tfl secondary_tfldisable_color_jitter img_size_min aa_params final_tflr+r+r,transforms_imagenet_trainAsA                    rcrop_pct crop_modecrop_border_pixelsc sptt|ttfrt|dksJtfdd|D} n t|} | | f} g}|r5|t|g7}| rB|t| | |dg7}nV|dkrW|t j | t |dt |g7}nA|dkrrdd|D}|t | |d d t||d g7}n&| d | d kr|t j | d t |dg7}n|t | g7}|t |g7}|r|tg7}n|s|tg7}n|tt jt|t|dg7}| r|t| dg7}t |S)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 csg|] }t|qSr+)mathfloorrXrr+r,r[7sz,transforms_imagenet_eval..)rDrErsquashr"bordercSsg|]}td|qSrT)rW)rYvr+r+r,r[RsrJ)rlongest)fillrr1r#rj)rrnrkrprtrrrrrr$r r%rrrrr&r'r(rr))rrrrrrrrr rCrDrErF scale_sizer*rr+rr,transforms_imagenet_evals\#    r input_size is_trainingno_augtf_preprocessingc"CsTt|ttfr|dd}n|}|r)|r)|rJdddlm} | ||| d}!|!S|r?|r?|r3Jdt|| ||||d}!|!S|rt|fid |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|| |||||||||||d" }!|!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)rrrrr r5r3r4r6r7r8r9r;r<r=rrrr>r?r@rArr rBrCrDrErFz>Separate transforms not supported for validation preprocessing) rrrrrrrr rCrDrErF)rnrkrptimm.data.tf_preprocessingrr-rr)"rrrr5r3r4r6r7r8r9r;r<r=rrrr>r?r@rArrrrrr rBrCrDrErFrr transformr+r+r,create_transformysD  > 6     - r))__doc__rtypingrrrr' torchvisionrtimm.data.constantsrrrtimm.data.auto_augmentr r r timm.data.transformsr r rrrrrrrtimm.data.naflex_transformsrrrtimm.data.random_erasingrrrroruboolr-rrrr+r+r+r,s ,    .        Q       j