kh}?&ddlZddlZddlZddlmZmZmZmZmZm Z m Z ddl Z ddl Z ddlmZddlmZmZddlmZmZddlmZddlmZmZdd lmZmZm Z m!Z!m"Z"m#Z#Gd d eZ$Gd d eZ%Gdde%Z&Gdde%Z'GddeZ(y)N)AnyCallableDictListOptionalSequenceUnion)one_hot) tree_flattentree_unflatten) transforms tv_tensors) functional)_RandomApplyTransform Transform)_check_sequence_input_parse_labels_getterhas_anyis_pure_tensor query_chw query_sizec eZdZdZej Zdeee fffd Z dde de e de e de de f fd Zd ed e d e d e de f fd Zdee deee ffdZd e deee fde fdZxZS) RandomErasingaNRandomly select a rectangle region in the input image or video and erase its pixels. This transform does not support PIL Image. 'Random Erasing Data Augmentation' by Zhong et al. See https://arxiv.org/abs/1708.04896 Args: p (float, optional): probability that the random erasing operation will be performed. scale (tuple of float, optional): range of proportion of erased area against input image. ratio (tuple of float, optional): range of aspect ratio of erased area. value (number or tuple of numbers): erasing value. Default is 0. If a single int, it is used to erase all pixels. If a tuple of length 3, it is used to erase R, G, B channels respectively. If a str of 'random', erasing each pixel with random values. inplace (bool, optional): boolean to make this transform inplace. Default set to False. Returns: Erased input. Example: >>> from torchvision.transforms import v2 as transforms >>> >>> transform = transforms.Compose([ >>> transforms.RandomHorizontalFlip(), >>> transforms.PILToTensor(), >>> transforms.ConvertImageDtype(torch.float), >>> transforms.Normalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225)), >>> transforms.RandomErasing(), >>> ]) returncptt| |jdS|jS)Nrandom)value)dictsuper _extract_params_for_v1_transformr)self __class__s ^/var/www/teggl/fontify/venv/lib/python3.12/site-packages/torchvision/transforms/v2/_augment.pyr!z.RandomErasing._extract_params_for_v1_transform2s8 G 4 6"jj0(  6:jj  pscaleratiorinplacecbt||t|tjt t tfs tdt|t r|dk7r tdt|ts tdt|ts td|d|dkDs |d|dkDrtjd |ddks|ddkDr td ||_ ||_t|tt frt!|g|_nTt|t rd|_n? ?  ec5\ *,DJ s #DJ e} -,12q%(2DJDJ ))ELL$<= 3s F,rinptargskwargsct|tjtjfrAt j t |jdt |jdt|$||g|i|S)Nz:() is currently passing through inputs of type tv_tensors.z(. This will likely change in the future.) r,r BoundingBoxesMaskr4r5type__name__r _call_kernel)r"rr=r>r?r#s r$rEzRandomErasing._call_kernel[st dZ55zG H MM:&&'("4j1122Z\ w#JFtFvFFr% flat_inputsc t|\}}}|j(t|jd|fvrtd|d||z}|j}t dD]}|t jdj|jd|jdjz}t jt jdj|d|dj} tttj|| z} tttj|| z } | |kr| |ks|j7t j|| | gt j j#} n't j$|jddddf} t j&d|| z dzdj} t j&d|| z dzdj}ndd||df\} }} } } t)| || | | S) Nrz@If value is a sequence, it should have either a single value or z (number of inpt channels) r)dtypersize)ijhwr<)rrlenr3r;ranger8emptyuniform_r'itemexpr6roundmathsqrtfloat32normal_r:randintr)r"rFimg_cimg_himg_warea log_ratio_ erase_area aspect_ratiorOrPr<rMrNs r$ make_paramszRandomErasing.make_paramscs' 4ue :: !3tzz?q%j+HRSXRYYst u}OO r 5A A 7 7 1 tzzRS} U Z Z \\J 99 A''aLaL df E$))J$=>?@AE$))J$=>?@AI!e)zz!KK1 U]]CKKMLL,Qd]; aQT:??AA aQT:??AA + 5.q%4MAq!Qa1Q!,,r%paramscp|d0|jtj|fi|d|ji}|S)Nr<r))rEFeraser)r"r=rfs r$ transformzRandomErasing.transforms7 #; "$4$$QWWdSfSdllSD r%)?)g{Gz?gQ?)g333333?gffffff @gF)rD __module__ __qualname____doc__ _transformsr_v1_transform_clsrr/rr!r7rboolr+rrErrerk __classcell__r#s@r$rrs<$11 $sCx. !-!+ !> !>!> !>  !>  !>FGxGsG3GRUGZ]G$-tCy$-T#s(^$-Lc4S>cr%rceZdZdddddedeeddffdZd Zd ejd efd Z d ejdedejfdZ xZ S)_BaseMixUpCutMix?Ndefault)alpha num_classes labels_getterryrzrc t|t||_tj j t j|gt j|g|_||_ t||_ yN) r r+r7ryr8 distributionsBetar:_distrzr_labels_getter)r"ryrzr{r#s r$r+z_BaseMixUpCutMix.__init__s` 5\ ((--ellE7.CU\\SXRYEZ[ &2=Ar%c l t|dkDr|n|d}t|\}}|j|}t|tj j t jt jr!tt|jd|j| t tjstdt d j dvrtd j"d j dk(rQ|j$E j"d |j$k7r)td j"d d |j$d  j dk(r|j$ td  j"dd|j't)||Dcgc] \}}|s | c}}}d|t+ fdt-|D<t)||Dcgc]\}}|r|j/||n|}}}t1||Scc}}wcc}}w)Nrrz9() does not support PIL images, bounding boxes and masks.z%The labels must be a tensor, but got z instead.)rzlabels should be index based with shape (batch_size,) or probability based with shape (batch_size, num_classes), but got a tensor of shape rzYWhen passing 2D labels, the number of elements in last dimension must match num_classes:  != z$. You can Leave num_classes to None.z=num_classes must be passed if the labels are index-based (1D))labels batch_sizeTc32K|]\}}|us |ywr}).0idxr=rs r$ z+_BaseMixUpCutMix.forward..s!`)#tQUY_Q_#!`s )rQr _needs_transform_listrPILImagerrArBr3rCrDrr,r8Tensorndimshaperzrezipnext enumeraterkr ) r"inputsrFspecneeds_transform_listr=needs_transformrf flat_outputsrs @r$forwardz_BaseMixUpCutMix.forwards+v;?q (0 T#99+F ; 1I1I:?? [T 3 344mno o$$V,&%,,/DT&\NR[\] ] ;;f $--3\\N)E  ;;!  0 0 <bAQUYUeUeAeT<<#$D)9)9(:;56  ;;!  0 0 8\] ] ,,q/ 58FZ5[o14_no eiT!`y7M!``a,/{`_. .. note:: This transform is meant to be used on **batches** of samples, not individual images. See :ref:`sphx_glr_auto_examples_transforms_plot_cutmix_mixup.py` for detailed usage examples. The sample pairing is deterministic and done by matching consecutive samples in the batch, so the batch needs to be shuffled (this is an implementation detail, not a guaranteed convention.) In the input, the labels are expected to be a tensor of shape ``(batch_size,)``. They will be transformed into a tensor of shape ``(batch_size, num_classes)``. Args: alpha (float, optional): hyperparameter of the Beta distribution used for mixup. Default is 1. num_classes (int, optional): number of classes in the batch. Used for one-hot-encoding. Can be None only if the labels are already one-hot-encoded. labels_getter (callable or "default", optional): indicates how to identify the labels in the input. By default, this will pick the second parameter as the labels if it's a tensor. This covers the most common scenario where this transform is called as ``MixUp()(imgs_batch, labels_batch)``. It can also be a callable that takes the same input as the transform, and returns the labels. rFrc^tt|jjdS)Nrr)rr7rsample)r"rFs r$rezMixUp.make_paramss!djj//3455r%r=rfc|d}||dur|j||St|tjtjfs t |r|j ||d|jddjd|z j|j|}t|tjtjfrtj|| }|S|S) Nrrrrrrrrwlike) rr,rrrrrrrrrwrap)r"r=rfroutputs r$rkzMixUp.transformsUm 6(# #$$Ts$3 3 z//1A1AB C~VZG[  & &t|8L & MYYq!_))#)499$((3-HF$!1!1:3C3C DE#d;MKr%N rDrmrnrorrrr/rerkrr%r$rrsH46tCy6T#s(^6c4S>cr%rcNeZdZdZdeedeeeffdZdedeeefdefdZ y) CutMixaApply CutMix to the provided batch of images and labels. Paper: `CutMix: Regularization Strategy to Train Strong Classifiers with Localizable Features `_. .. note:: This transform is meant to be used on **batches** of samples, not individual images. See :ref:`sphx_glr_auto_examples_transforms_plot_cutmix_mixup.py` for detailed usage examples. The sample pairing is deterministic and done by matching consecutive samples in the batch, so the batch needs to be shuffled (this is an implementation detail, not a guaranteed convention.) In the input, the labels are expected to be a tensor of shape ``(batch_size,)``. They will be transformed into a tensor of shape ``(batch_size, num_classes)``. Args: alpha (float, optional): hyperparameter of the Beta distribution used for mixup. Default is 1. num_classes (int, optional): number of classes in the batch. Used for one-hot-encoding. Can be None only if the labels are already one-hot-encoded. labels_getter (callable or "default", optional): indicates how to identify the labels in the input. By default, this will pick the second parameter as the labels if it's a tensor. This covers the most common scenario where this transform is called as ``CutMix()(imgs_batch, labels_batch)``. It can also be a callable that takes the same input as the transform, and returns the labels. rFrct|jjd}t|\}}t j |d}t j |d}dt jd|z z}t||z}t||z} tt j||z d} tt j|| z d} tt j||z|} tt j|| z|} | | | | f}td| | z | | z z||zz z }t|| S) NrrJrKrlrwr)min)max)box lam_adjusted) r7rrrr8r\rXrYr6clampr)r"rFrHWr_xr_yrr_w_halfr_h_halfx1y1x2y2rrs r$rezCutMix.make_params'sDJJ%%b)*+&1mmAD)mmAD) $))C#I& &q1u:q1u: S8^3 4 S8^3 4 S8^3 4 S8^3 42r2SBGR#8AE#BBC ,77r%r=rfc||dur|j||dSt|tjtjfs t |r|j ||d|d\}}}}|jdd}|j}|d ||||f|d ||||f<t|tjtjfrtj|| }|S|S) Nrrrrrrrr.r) rr,rrrrrrcloner) r"r=rfrrrrrolledrs r$rkzCutMix.transform=s 6(# #$$Tvn/E$F F z//1A1AB C~VZG[  & &t|8L & M#E]NBBYYq!_FZZ\F(.sBrE2b5/@(AF32r"u$ %$!1!1:3C3C DE#d;MKr%Nrrr%r$rr sH68tCy8T#s(^8,c4S>cr%rcveZdZdZdeeeefffd Zdee de e e ffdZ de de e e fde fd Z xZS) JPEGa>Apply JPEG compression and decompression to the given images. If the input is a :class:`torch.Tensor`, it is expected to be of dtype uint8, on CPU, and have [..., 3 or 1, H, W] shape, where ... means an arbitrary number of leading dimensions. Args: quality (sequence or number): JPEG quality, from 1 to 100. Lower means more compression. If quality is a sequence like (min, max), it specifies the range of JPEG quality to randomly select from (inclusive of both ends). Returns: image with JPEG compression. qualitycHt|t|tr t|tr t d||g}nt |ddd|dcxkr|dcxkrdkr)nn&t|dtrt|dtstd|||_y) Nzquality can't be boolr)r) req_sizesrrdz7quality must be an integer from 1 to 100, got quality =) r r+r,r6rrr2rr3r)r"rr#s r$r+z JPEG.__init__`s  gs #'4( 788(G !'9 EWQZ471:44GAJPS9TYcdklmdnpsYtWg[YZ Z r%rFrctj|jd|jddzdj}t |S)Nrrrr)r8r\rrUr)r"rFrs r$rezJPEG.make_paramsns=-- Qa11DbINNPG$$r%r=rfcL|jtj||dS)Nrr)rErhjpegrjs r$rkzJPEG.transformrs#  vi7H IIr%)rDrmrnror r6rr+rrrr/rerkrsrts@r$rrPsg  c8C=&8 9 %tCy%T#s(^%JcJ4S>JcJr%r))rXr-r4typingrrrrrrr PIL.Imagerr8torch.nn.functionalr torch.utils._pytreer r torchvisionr rprtorchvision.transforms.v2rrh _transformrr_utilsrrrrrrrrvrrrrr%r$rs GGG '<=58oo|)|~HEyHEV- -`B BJ#J9#Jr%