bCi`ZSSKJr SSKJr SSKJrJr SSKJrJ r J r J r SSK r SSKJrJrJr SSKJrJrJrJrJr SS KJrJrJrJrJrJrJrJ r J!r!J"r"J#r# SS K$J%r% SS K&J'r'J(r(J)r)J*r*J+r+J,r, SS K-J.r. \+"5(aSS KJ/r/ \)"5(aSSK0r0\*"5(a SSK1J2r3 SSKJ4r4 OSr4\,Rj"\65r7\"SS9SSSSSSSSSSS\Rp4 S\ \9S\ \:S\ \9S\ \ \:\;\:4S\ \ \:\;\:4S\ \9S\ \S\ \9S\ \S\ SS\ \ \<\'4S\4Sjj5r=S3S S!S"\ \>S#S!4S$jjr?S%\\S#\;\4S&jr@S'\;S!S#\A\>S(44S)jrBS*\ \ RS!4S+\>S#\;\ \ RS!44S,jrD"S-S.\ S/S09rE\("S1S2\55rFg)4)Iterable)deepcopy) lru_cachepartial)AnyOptional TypedDictUnionN)BaseImageProcessor BatchFeature get_size_dict)convert_to_rgbget_resize_output_image_sizeget_size_with_aspect_ratiogroup_images_by_shapereorder_images) ChannelDimension ImageInput ImageTypeSizeDictget_image_size#get_image_size_for_max_height_widthget_image_typeinfer_channel_dimension_formatmake_flat_list_of_imagesvalidate_kwargsvalidate_preprocess_arguments)Unpack) TensorTypeauto_docstringis_torch_availableis_torchvision_availableis_vision_availablelogging)is_rocm_platform)PILImageResampling) functional)pil_torch_interpolation_mapping maxsize do_rescalerescale_factor do_normalize image_mean image_stddo_center_crop crop_size do_resizesize interpolationF.InterpolationModereturn_tensors data_formatc [UUUUUUUUUU S9 U bU S:wa [S5eU [R:wa [S5eg)z Checks validity of typically used arguments in an `ImageProcessorFast` `preprocess` method. Raises `ValueError` if arguments incompatibility is caught. ) r-r.r/r0r1r2r3r4r5r6Nptz6Only returning PyTorch tensors is currently supported.z6Only channel first data format is currently supported.)r ValueErrorrFIRST) r-r.r/r0r1r2r3r4r5r6r8r9s b/home/james-whalen/.local/lib/python3.13/site-packages/transformers/image_processing_utils_fast.py"validate_fast_preprocess_argumentsr?Ise&"%!% # !n&<QRR&,,,QRR-tensor torch.TensoraxisreturnclUcUR5$URUS9$![a Us$f=f)z> Squeezes a tensor, but only if the axis specified has dim 1. )rC)squeezer<)rArCs r> safe_squeezerGps@ |~~~~4~((  s $ 33valuescP[U6Vs/sHn[U5PM sn$s snf)zG Return the maximum value across all indices of an iterable of values. )zipmax)rHvalues_is r>max_across_indicesrM}s$+.v, 7,hCM, 77 7s#images.cb[UVs/sHoRPM sn5up#nX44$s snf)z@ Get the maximum height and width across all images in a batch. )rMshape)rNimg_ max_height max_widths r>get_max_height_widthrUs3 22O992OPA9  ""3Ps,image patch_sizec/n[U[RS9up4[SX15H9n[SXA5H&nUSS2XUU-2XfU-24nUR U5 M( M; U$)a Divides an image into patches of a specified size. Args: image (`Union[np.array, "torch.Tensor"]`): The input image. patch_size (`int`): The size of each patch. Returns: list: A list of Union[np.array, "torch.Tensor"] representing the patches. ) channel_dimrN)rrr=rangeappend)rVrWpatchesheightwidthijpatchs r>divide_to_patchesrbsrG"56F6L6LMMF 1f )q%,A!QZ/^1CCDE NN5 !-* Nr@c\rSrSr%\\\S'\\\\ 4\S'\\\S'\\ S\S'\\\S'\\\\ 4\S'\\\S '\\ \ \ 4\S '\\\S '\\ \ \ \ 4\S '\\ \ \ \ 4\S '\\\S'\\\\ 4\S'\\\S'\\ \\ 4\S'\\\S'\\ \\4\S'\S\S'\\\S'Srg)DefaultFastImageProcessorKwargsr4r5default_to_square)r'r7resampler2r3r-r.r/r0r1do_padpad_sizedo_convert_rgbr8r9input_data_format torch.devicedevicedisable_groupingN)__name__ __module__ __qualname____firstlineno__rbool__annotations__dictstrintr floatlistr r__static_attributes__ror@r>rdrds#~ 4S> ""~%uHIJJTN"S#X''U3:.//4. ud5k1233eT%[0122 TNtCH~&&TN"U3 ?344*++c+;&; <== ^ $$tn$r@rdF)totalc!^\rSrSrSrSrSrSrSrSr Sr Sr Sr Sr SrSrSrSrSr\R(rSrSrS/r\rSrS\\4U4Sjjr\S\4S j5r SCS S S \!S \"\#S\"\$S\S\"\SS 4Sjjr%SDSS S\!S\"SS\SS 4 Sjjr&\'SDSS S\(\#\#4S\"SS\SS 4 Sjj5r)SS S\*SS 4Sjr+SS S\,\*\-\*4S\,\*\-\*4SS 4Sjr.\/"SS 9SES!\"\S"\"\,\*\0\*4S#\"\,\*\0\*4S$\"\S%\"\*S&\"S'S\(4S(jj5r1S S S$\S%\*S!\S"\,\*\0\*4S#\,\*\0\*4SS 4S)jr2SS S\!SS 4S*jr3S\4S\44S+jr5S\64S,jr7SFS \4S-\#S\44S.jjr8SGS\4S/\"\S0\"\,\$\4S&\"S'SS 4 S1jjr9SHS \4S/\"\S0\"\,\$\4S&\"S'S-\#S\0S 4 S2jjr:SIS\"\!S3\"\!S \"\!S4\"\S"\"\,\*\0\*4S#\"\,\*\0\*4S5\"\S\64S6jjr;SJS$\"\S%\"\*S!\"\S"\"\,\*\(\*4S#\"\,\*\(\*4S7\"\S\"\!S8\"\S3\"\!S\"SS9\"\,\$\<4S5\"\4S:jjr=S \4S\\S\>4S;jr?\@S \4S\\S\>4S<j5rASS=.S \4S/\S0\S&\"\,\$S'4S\\S\>4 S>jjrBS \0S S7\S\!S\"SS8\S3\!S$\S%\*S!\S"\"\,\*\0\*4S#\"\,\*\0\*4S?\"\S \"\!S\"\S9\"\,\$\<4S\>4 S@jrCU4SAjrDSBrEU=rF$)KBaseImageProcessorFastNTgp? pixel_valueskwargsc >[TU]"S0UD6 URU5nURSUR5nUb#[ X!RSUR 5S9OSUlURSUR5nUb [ USS9OSUlURSUR5nUb [ USS9OSUlURRHEnURUS5nUb[XU5 M&[X[[XS555 MG [URRR55Ulg) Nr5rfr5rfr3 param_namerir5rro)super__init__filter_out_unused_kwargspopr5rrfr3ri valid_kwargsrusetattrrgetattrrzkeys_valid_kwargs_names)selfrr5r3rikeykwarg __class__s r>rBaseImageProcessorFast.__init__s, "6"..v6zz&$)), tzzBUW[WmWm7n o JJ{DNN; MVMby[Ihl::j$--8OWOc 8 Kim $$44CJJsD)E 5)8GDt,D#EF 5$((9(9(I(I(N(N(P#Q r@rDcg)zf `bool`: Whether or not this image processor is a fast processor (backed by PyTorch and TorchVision). Tro)rs r>is_fastBaseImageProcessorFast.is_fasts r@rNrBri fill_value padding_mode return_maskrnc UbJUR(aUR(d[SUS35eURUR4nO [U5n[ XS9up0n 0n UR 5HupU R SSnUSUS- nUSUS- nUS:dUS:a[SUS US35eX:waSSUU4n[R"U UX4S 9n XU 'U(dMv[R"U [RS 9S SSS2SS24nSUS SUS2SUS24'UX'M [X5nU(a[X5nUU4$U$) a Pads images to `(pad_size["height"], pad_size["width"])` or to the largest size in the batch. Args: images (`torch.Tensor`): Images to pad. pad_size (`SizeDict`, *optional*): Dictionary in the format `{"height": int, "width": int}` specifying the size of the output image. fill_value (`int`, *optional*, defaults to `0`): The constant value used to fill the padded area. padding_mode (`str`, *optional*, defaults to "constant"): The padding mode to use. Can be any of the modes supported by `torch.nn.functional.pad` (e.g. constant, reflection, replication). return_mask (`bool`, *optional*, defaults to `False`): Whether to return a pixel mask to denote padded regions. disable_grouping (`bool`, *optional*, defaults to `False`): Whether to disable grouping of images by size. Returns: `torch.Tensor`: The resized image. NzCPad size must contain 'height' and 'width' keys only. Got pad_size=.rnrr zrPadding dimensions are negative. Please make sure that the `pad_size` is larger than the image size. Got pad_size=z , image_size=)fillrdtype.) r]r^r<rUritemsrPFpadtorch zeros_likeint64r)rrNrirrrrnrgrouped_imagesgrouped_images_indexprocessed_images_groupedprocessed_masks_groupedrPstacked_images image_sizepadding_height padding_widthpadding stacked_masksprocessed_imagesprocessed_maskss r>rBaseImageProcessorFast.pads>  OO #fgofppq!rss 8H+F3H/DV/o,#% "$%3%9%9%; !E'--bc2J%a[:a=8N$QK*Q-7M!]Q%6 008zzlRSU%a?!"~wZ!k.'.%&<(**BY ,-D[O#_4 4r@rVr5r6r7 antialiasc FUbUO[RRnUR(aDUR(a3[ UR 5SSURUR5nOUR(a%[UURS[RS9nOUR(aDUR(a3[UR 5SSURUR5nOJUR(a*UR(aURUR4nO[SUS35e[ R"R%5(a!['5(aUR)XX45$[R*"XX4S9$)a Resize an image to `(size["height"], size["width"])`. Args: image (`torch.Tensor`): Image to resize. size (`SizeDict`): Dictionary in the format `{"height": int, "width": int}` specifying the size of the output image. interpolation (`InterpolationMode`, *optional*, defaults to `InterpolationMode.BILINEAR`): `InterpolationMode` filter to use when resizing the image e.g. `InterpolationMode.BICUBIC`. Returns: `torch.Tensor`: The resized image. NrF)r5rfrkzjSize must contain 'height' and 'width' keys, or 'max_height' and 'max_width', or 'shortest_edge' key. Got rr6r)rInterpolationModeBILINEAR shortest_edge longest_edgerr5rrr=rSrTrr]r^r<rcompiler is_compilingr&compile_friendly_resizeresize)rrVr5r6rrnew_sizes r>rBaseImageProcessorFast.resize7s<,*7)B H[H[HdHd   $"3"32 RS!""!!H   3''"'"2"8"8 H __:5::<;Ldoo_c_m_mnH [[TZZ TZZ0H6  >> & & ( (-=-?-?//Z Zxx}ZZr@rcUR[R:XaUR5S- n[R "XX#S9nUS-n[R "US:SU5n[R "US:SU5nUR5R[R5nU$[R "XX#S9nU$)zk A wrapper around `F.resize` so that it is compatible with torch.compile when the image is a uint8 tensor. rr) rruint8ryrrwhereroundto)rVrr6rs r>r.BaseImageProcessorFast.compile_friendly_resizems ;;%++ %KKMC'EHHUM_ECKEKK S%8EKK 1e4EKKM$$U[[1E HHUM_E r@scalec X-$)z Rescale an image by a scale factor. image = image * scale. Args: image (`torch.Tensor`): Image to rescale. scale (`float`): The scaling factor to rescale pixel values by. Returns: `torch.Tensor`: The rescaled image. ro)rrVrrs r>rescaleBaseImageProcessorFast.rescales $}r@meanstdc 0[R"XU5$)a Normalize an image. image = (image - image_mean) / image_std. Args: image (`torch.Tensor`): Image to normalize. mean (`torch.Tensor`, `float` or `Iterable[float]`): Image mean to use for normalization. std (`torch.Tensor`, `float` or `Iterable[float]`): Image standard deviation to use for normalization. Returns: `torch.Tensor`: The normalized image. )r normalize)rrVrrrs r>r BaseImageProcessorFast.normalizes*{{5,,r@r*r+r/r0r1r-r.rmrlcU(a=U(a6[R"X&S9SU- -n[R"X6S9SU- -nSnX#U4$)Nrmg?F)rrA)rr/r0r1r-r.rms r>!_fuse_mean_std_and_rescale_factor8BaseImageProcessorFast._fuse_mean_std_and_rescale_factorsI ,j@C.DXYJ Y>#BVWIJj00r@c URUUUUUURS9upVnU(a/URUR[R S9XV5nU$U(aUR X5nU$)z Rescale and normalize images. )r/r0r1r-r.rmr)rrmrrrfloat32r)rrNr-r.r/r0r1s r>rescale_and_normalize,BaseImageProcessorFast.rescale_and_normalizesx-1,R,R%!!)== -S- ) z ^^FIIEMMI$BJZF \\&9F r@c URb URc[SUR535eURSSupEURURpvXu:dXd:anXu:aXu- S-OSXd:aXd- S-OSXu:a Xu- S-S-OSXd:a Xd- S-S-OS/n[ R "XSS9nURSSupEXu:XaXd:XaU$[XF- S- 5n [XW- S- 5n [ R"XXU5$) a Note: override torchvision's center_crop to have the same behavior as the slow processor. Center crop an image to `(size["height"], size["width"])`. If the input size is smaller than `crop_size` along any edge, the image is padded with 0's and then center cropped. Args: image (`"torch.Tensor"`): Image to center crop. size (`dict[str, int]`): Size of the output image. Returns: `torch.Tensor`: The center cropped image. Nz=The size dictionary must have keys 'height' and 'width'. Got rrr )rg@) r]r^r<rrPrrrxcrop) rrVr5r image_height image_width crop_height crop_width padding_ltrbcrop_top crop_lefts r> center_crop"BaseImageProcessorFast.center_crops+( ;; $**"4\]a]f]f]h\ijk k$)KK$4! "&++tzzZ  #{'A3=3K)a/QR5@5O+1UV7A7O)A-!3UV9D9S+a/A5YZ L EE%A6E(- BC(8 %L([-H  2c9:1S89 vveyzJJr@c[U5$)z Converts an image to RGB format. Only converts if the image is of type PIL.Image.Image, otherwise returns the image as is. Args: image (ImageInput): The image to convert. Returns: ImageInput: The converted image. )r)rrVs r>r%BaseImageProcessorFast.convert_to_rgbse$$r@cURcU$URH4nX!;dM [RSUS35 URU5 M6 U$)z: Filter out the unused kwargs from the kwargs dictionary. z!This processor does not use the `z ` parameter. It will be ignored.) unused_kwargslogger warning_oncer)rr kwarg_names r>r/BaseImageProcessorFast.filter_out_unused_kwargssW    %M,,J###&G |Ss$tu :&- r@expected_ndimsc6URU5n[XS9$)z Prepare the images structure for processing. Args: images (`ImageInput`): The input images to process. Returns: `ImageInput`: The images with a valid nesting. r) fetch_imagesr)rrNrs r>_prepare_images_structure0BaseImageProcessorFast._prepare_images_structure$s ""6*'NNr@rjrkc[U5nU[R[R[R4;a[ SU35eU(aUR U5nU[R:Xa[R"U5nO8U[R:Xa$[R"U5R5nURS:XaURS5nUc [U5nU[R :Xa!UR#SSS5R5nUbUR%U5nU$)NzUnsupported input image type rrr )rrPILTORCHNUMPYr<rr pil_to_tensorr from_numpy contiguousndim unsqueezerrLASTpermuter)rrVrjrkrm image_types r>_process_image%BaseImageProcessorFast._process_image7s$E* immY__iooN Nu E   0 5 5 5MM!Q*557E  HHV$E r@c \URXS9n[URX#US9n[U5S:=(a [ US[ [ 45nU(a+UVV s/sHoV s/sH o"U 5PM sn PM n nn U $UV s/sH o"U 5PM n n U $s sn fs sn nfs sn f)aZ Prepare image-like inputs for processing. Args: images (`ImageInput`): The image-like inputs to process. do_convert_rgb (`bool`, *optional*): Whether to convert the images to RGB. input_data_format (`str` or `ChannelDimension`, *optional*): The input data format of the images. device (`torch.device`, *optional*): The device to put the processed images on. expected_ndims (`int`, *optional*): The expected number of dimensions for the images. (can be 2 for segmentation maps etc.) Returns: List[`torch.Tensor`]: The processed images. rrjrkrmr)rrrlen isinstancerztuple) rrNrjrkrmrprocess_image_partialhas_nested_structure nested_listrQrs r>_prepare_image_like_inputs1BaseImageProcessorFast._prepare_image_like_inputs]s8///V '   lr!  #6{QW:fQi$PU3W gmngmXc{ S{!6s!;{ Sgm n GMMfs 5c :f M !TnMs B#(B 9B# B)B#r3rfr9c Uc0nUb[S 0[XS9D6nUb[S 0[USS9D6nUb[S 0[USS9D6n[U[5(a [ U5n[U[5(a [ U5nUc[ R nXS'X(S'X8S'XXS'XhS'XxS 'URS 5n [U [[45(a [U OU US 'U$) z Update kwargs that need further processing before being validated Can be overridden by subclasses to customize the processing of kwargs. rr3rrirr5r0r1r9rgr6ro) rrrrzrrr=rr'rxr)) rr5r3rirfr0r1r9rrgs r>_further_process_kwargs.BaseImageProcessorFast._further_process_kwargss >F  \m[\D   T={#STI  V-X*"UVH j$ ' 'z*J i & &i(I  *00Kv'{%z)|'{ +} ::j)9CHOacfNg9h9h +H 5nv  r@r4r2r8c  ,[UUUUUUUUU U U U S9 g)z0 validate the kwargs for the preprocess method. ) r-r.r/r0r1r4r5r2r3r6r8r9N)r?)rr-r.r/r0r1r4r5r2r3r6r8r9rs r>_validate_preprocess_kwargs2BaseImageProcessorFast._validate_preprocess_kwargss0& +!)%!)')# r@c.UR"U/UQ70UD6$N) preprocess)rrNargsrs r>__call__BaseImageProcessorFast.__call__sv7777r@c [UR5URS9 URHnURU[ XS55 M! UR S5nUR S5nUR S5nUR "S0UD6nUR"S0UD6 UR S5 UR"U/UQ7XVUS.UD6$)N)captured_kwargsvalid_processor_keysrjrkrmr9rro) rrr setdefaultrrrr_preprocess_image_like_inputs)rrNrrrrjrkrms r>r!BaseImageProcessorFast.preprocesss  DLdLde22J   j'$D*I J3 $45"JJ':;H%--77 ((262  =!11   *8fl pv  r@rcNURXX4S9nUR"U/UQ70UD6$)z Preprocess image-like inputs. To be overridden by subclasses when image-like inputs other than images should be processed. It can be used for segmentation maps, depth maps, etc. )rNrjrkrm)r  _preprocess)rrNrjrkrmrrs r>r4BaseImageProcessorFast._preprocess_image_like_inputss<00L]1 8888r@rhc [XS9unn0nUR5H"unnU(aURUX4S9nUUU'M$ [UU5n[UUS9unn0nUR5H8unnU(aUR UU5nUR UXxXU 5nUUU'M: [UU5nU (aUR UXS9nU(a[R"USS9OUn[SU0US9$)Nr)rVr5r6)rirnr)dimr)data tensor_type) rrrrrrrrstackr )rrNr4r5r6r2r3r-r.r/r0r1rhrirnr8rrrresized_images_groupedrPrresized_imagesrrs r>r "BaseImageProcessorFast._preprocess s&(0EV/o,,!#%3%9%9%; !E>!%>!j,: "5 )&<((>@TU0E^fv/w,,#% %3%9%9%; !E>!%!1!1.)!L!77 LV_N/= $U +&<**BDXY #xx(88xo CQ5;;'7Q?Wg.2B!CQ_``r@cl>[TU]5nURSS5 URSS5 U$)N_valid_processor_keysr)rto_dictr)r encoder_dictrs r>r,BaseImageProcessorFast.to_dict=s7w( 0$7.5r@)rr3rir5)NrconstantFF)NT)NNNNNN))NNN)NNNr0)NNNNNNN) NNNNNNNNNNNN)Grprqrrrsrgr0r1r5rfr3r4r2rhrir-r.r/rjr8rr=r9rkrmmodel_input_namesrdrrrrpropertyrtrrrrxrwrr staticmethodrrryrr rrrrzrrrrrrvrrrr rr rr rr!rrr r,r{ __classcell__)rs@r>r~r~sHJI DIIN FHJNLNN"((K F'(2LMR(G!HR0"$%&0!+0B B B SM B sm B  B #4.B  B P:> 4[4[4[ 56 4[  4[ 4[l:> S/ 56   0  (--E8E?*+-5(5/) * -  -.r(,:>9=%)*.+/1tn1U5$u+#5671E%e"456 1 TN 1 ! 1(1 11     %e,- U +, 8'K'K'K  'KR%% % t   OOO  O,*.DH+/ $$!$$E#/?*?$@A $ ( $  $R*.DH+/ * * !* $E#/?*?$@A * ( *  *  n * \$((,'+,0:>9=26.x .H%.8$ . $D> . U5$u+#567 .E%e"456../. .d&**.'+;?:>$(#')-(,9=;?26  TN  !  tn  U5%,#678  E%u"567  D>  x   !  H%   56  !sJ!78  ./  D8z8FCbe jv  @8< 99 9 , 9 sN234 9899 9(.a^$.a.a .a  56 .a  .a.a.a.a.aU5$u+#567.aE%e"456.a.a8$.a#4..a !sJ!78!.a$ %.a`r@r~r)Gcollections.abcrcopyr functoolsrrtypingrrr r numpynpimage_processing_utilsr r rimage_transformsrrrrr image_utilsrrrrrrrrrrrprocessing_utilsrutilsr r!r"r#r$r%utils.import_utilsr&r'rtorchvision.transforms.v2r(rr) get_loggerrprr=rtryrzrwr?rxrGrMrrUndarrayrbrdr~ror@r>rDsv%(22SS    %1/9<'+#   H % 2!%&*#'6:59%)$( $#597;$4$:$:#S#SUO#S4.#Sud5k123 #S eT%[012 #S TN #S!#S~#S 8 #S12#SU3 ?34#S"#S#SL  x}  8x}8c8#n!5#%S/# ^+ ,:= % N* +,0%iu%,D /D D r@