+ j\U^RIHt^RIt^RIHt^RIHuHt^RIH t ^RI H t ^RI H t HtHtHt^RIHt^RIHt^RIHt^R IHtHt^R IHtHt^R IHt^R IH t H!t!^R I"H#t#^RI$H%t%^RI&H't'H(t(H)t)^RI*H+t+H,t,^RI-H.t.^RI/H0t0^RI1H2t2^RI3H4t4^RI5H6t6](!RR7] !RR] ]444t7](!RR7] !RR]444t8](].!R8R7!RR]444t9!RR ]44t:!R!R"]Pv4t<!R#R$]Pv4t=!R%R&]Pv4t>!R'R(]4t?!R)R*]Pv4t@](!R+R,]644tA!R-R.]A4tB](!R/R07!R1R2]]A44tC!R3R4]Pv4tD](!R5R07!R6R7]A44tE.R9OtFR#):)SequenceN)strict)ACT2FN)BackboneConfigMixin BackboneMixin%consolidate_backbone_kwargs_to_configfilter_output_hidden_states)PreTrainedConfig) BatchFeature)TorchvisionBackend)group_images_by_shapereorder_images)PILImageResamplingSizeDict)GradientCheckpointingLayer)BackboneOutputBaseModelOutputWithNoAttention)PreTrainedModel)Unpack)TransformersKwargsauto_docstringcan_return_tuple) TensorTypemerge_with_config_defaults)requires)capture_outputs) AutoConfig)PPLCNetConvLayer)PPOCRV5ServerDetPreTrainedModelzPaddlePaddle/UVDoc_safetensors) checkpointc^aa]tRt^0toRtRtRtRtRtRt R t ^t V3Rlt V3Rlt RtVtV;t#) UVDocBackboneConfigal resnet_head (`Sequence[list[int] | tuple[int, ...]]`, *optional*, defaults to `((3, 32), (32, 32))`): Configuration for the ResNet head layers in format [in_channels, out_channels]. resnet_configs (`Sequence[Sequence[tuple[int, int, int, bool] | list[int | bool]]]`, *optional*, defaults to `(((32, 32, 1, False), (32, 32, 3, False), (32, 32, 3, False)), ((32, 64, 1, True), (64, 64, 3, False), (64, 64, 3, False), (64, 64, 3, False)), ((64, 128, 1, True), (128, 128, 3, False), (128, 128, 3, False), (128, 128, 3, False), (128, 128, 3, False), (128, 128, 3, False)))`): Configuration for the ResNet stages in format [in_channels, out_channels, dilation_value, downsample]. stage_configs (Sequence[Sequence[tuple[int, ...] | list[int]]], *optional*, defaults to `(((128, 1),), ((128, 2),), ((128, 5),), ((128, 8),(128, 3),(128, 2),), ((128, 12), (128, 7), (128, 4),), ((128, 18), (128, 12), (128, 6),),)`): Configuration for the bridge module stages in format [in_channels, dilation_value]. Each inner sequence corresponds to a single bridge block, and the outer sequence groups blocks by bridge stage. uvdoc_backboneNc <VPUu.uFp\V4NK upVnR.\^\VP4^,4Uu.uFpRV 2NK up,VnVP VP RR4VP RR4R7\SV` !R/VBR#uupiuupi)stemstage out_indicesN out_features)r'r() stage_configslendepthsrange stage_names"set_output_features_output_indicespopsuper __post_init__)selfkwargsstagesidx __class__s&, h/home/luca/screen-checking/.venv/lib/python3.14/site-packages/transformers/models/uvdoc/modular_uvdoc.pyr2!UVDocBackboneConfig.__post_init__ws151C1CD1Cvs6{1CD "8aTM_M_I`cdId@e&f@ese}@e&ff // =$7fjjQ_aeFf 0  '' E&fs B7B<c<V^8dQh/S[S[,R,;R&S[S[,R,;R&S[S[S[,S[S[R3,,,;R&S[S[S[S[S[S[S[3,S[S[S[,,,,,;R&S[S[S[S[R3,S[S[,,,,;R&S[;R&#) N _out_features _out_indices. resnet_headresnet_configsr* kernel_size)liststrintrtuplebool)format __classdict__s"r8 __annotate__ UVDocBackboneConfig.__annotate__0s$9t#*%&s)d")'*$s)eCHo56+4XeCc4,?&@4d CS&STU5`HU38_tCy%@ABaJK)r,r.)) )rLrL)))rLrLFrLrLrKFrN))rL@rMTrOrOrKFrPrP))rOrMTrQrQrKFrRrRrRrR)))rQrM)rQr;))rQ))rQ)rQrKrS)rQ )rQ)rQ))rQrV)rQ)__name__ __module__ __qualname____firstlineno____doc__ model_typer<r=r>r?r*r@r2__annotate_func____static_attributes____classdictcell__ __classcell__r7rGs@@r8r"r"0sF "J&*M%)L:K YN,FM*K(OrJr"cfaa]tRt^toRtRtR]/tRtRt Rt ^t R t R t V3RltV3RltR tVtV;t#) UVDocConfiga padding_mode (`str`, *optional*, defaults to `"reflect"`): Padding mode for convolutional layers. Supported modes are `"reflect"`, `"constant"`, and `"replicate"`. kernel_size (`int`, *optional*, defaults to 5): Kernel size for convolutional layers in the backbone network. bridge_connector (`list[int] | tuple[int, ...]`, *optional*, defaults to `(128, 128)`): Configuration for the bridge connector in format [in_channels, out_channels]. out_point_positions2D (`Sequence[list[int] | tuple[int, ...]]`, *optional*, defaults to `((128, 32), (32, 2))`): Configuration for the output point positions 2D layer in format [in_channels, out_channels]. uvdocbackbone_configNprelureflectc j<\RRVPRR/VBwVnp\SV` !R/VBR#)rjdefault_config_typer#Nr))rrjr1r2)r3r4r7s&,r8r2UVDocConfig.__post_init__sG'L(  00( 0( ( $f ''rJc<V^8dQh/S[S[,R,;R&S[;R&S[;R&S[;R&S[S[,S[S[R3,,;R&S[S[S[,S[S[R3,,,;R&#) r;Nrj hidden_act padding_moder@.bridge_connectorout_point_positions2D)dictr rBrCrArDr)rFrGs"r8rHUVDocConfig.__annotate__s ,,t3:!$%&!'()*3i%S/1>+,$DIc3h$?@W-rJ)rj)rQrQ))rQrL)rLr;)r\r]r^r_r`rar sub_configsrjrqrrr@rsrtr2rbrcrdrerfs@@r8rhrhsG J$j1K6:OJ!LK4>CW(1rJrh)backendscta]tRt^toRtRtRRRR/t]Pt V3RlRlt R V3RlR llt R t Vt R #) UVDocImageProcessorTheightiwidthc<V^8dQhRS[R,RS[RS[RS[RS[RS[RS[S[S[,,R ,R S[S[S[,,R ,R S[R ,R S[S[,R ,R S[/ #)r;imagesz torch.Tensor do_resizesize do_rescalerescale_factor do_normalize image_meanN image_stddisable_groupingreturn_tensorsreturn)rArErfloatrBrr )rFrGs"r8rH UVDocImageProcessor.__annotate__s/ / ^$/ /  /  /  / / DK'$./ 4;&-/ +/ j(4//  / rJc \WR7wr/pV P4F-wppVPVWEWgV4pVR.RORR3,pVW&K/ \W4pVP 4p\VV R7wr/pV P4FCwppV'd1\ P !VVPVP3RRR7pVVV&KE \VV 4p\RVRV/V R.R7#) )r:NNNbilinearTrmode align_corners pixel_valuesoriginal_images)data tensor_typeskip_tensor_conversion)r;rMr) r itemsrescale_and_normalizercopyF interpolater{r|r )r3rrrrrrrrrrr4grouped_imagesgrouped_images_indexprocessed_images_groupedshapestacked_imagesrescale_and_normalize_imagesrinterpolated_images_groupedrs&&&&&&&&&&&, r8 _preprocessUVDocImageProcessor._preprocesss 0EV/o,#% %3%9%9%; !E>!77 LV_N,Ay!Q,>?N.< $ + &<(66N'e$6;;=/D (;K0 ,')#%3%9%9%; !E>!""$++tzz)Bcg"2@ ' . &<&&ACWX  ,0A?S&$5#6  rJc <V^8dQhRS[PRS[S[P,RS[RS[S[S[S[P3,,/#)r; predictionrscaler)torchTensorrArrurB)rFrGs"r8rHrsS55LL5ell+5 5 d3 $% & 5rJc \V4p\P!R \V4VPR7p.p\ V4EFwrgVP ^8XdVP^4pVPVP4pVPR,wr\P!WV^,W3RRR7p V P^^^^4p \P!W{RR7p V P^4P^^^4p W,p V PR .R7P\P RRR 7p VP#R V /4EK V#) aN Post-process document rectification predictions to convert them into rectified images. Args: prediction: Predicted 2D Bezier mesh coordinates, shape (B, 2, H, W) original_images: List of original input tensors, each of shape (C, H_i, W_i). Images may have different sizes. scale: Scaling factor for output images (default: 255.0) Returns: List of dictionaries containing rectified images. Each dictionary has: - "images": Rectified image tensor of shape (H, W, 3) with dtype torch.uint8 and BGR channel order (suitable for OpenCV visualization) )device:r;NNrTr)r)dimsF)dtype non_blockingrrr))rArfullrr enumeratendim unsqueezetorrrpermute grid_samplesqueezeflipuint8append)r3rrr image_listresultsioriginal_imageoriginal_heightoriginal_widthupsampled_meshrearranged_mesh rectifiedimages&&&& r8#post_process_document_rectification7UVDocImageProcessor.post_process_document_rectificationsF&/*  2uU|J4E4EF!*:!6 A""a'!/!9!9!!<+..z/@/@AN.<.B.B2.F +O]]q1u%%6" N-44Q1a@O nUYZI%%a(00Aq9EMEJJRDJ),,5;;TX],^E NNHe, -9"7<rJr)N)go@)r\r]r^r_rrrrBILINEARresamplerrrcrdrGs@r8rzrzsDJI c7C (D!**H/ / b555rJrzcFaa]tRtRtoRtRV3RlV3RllltRtVtV;t#)UVDocConvLayeriz>-0 6  6 %0  M2 rJ)rrrr)rMrrMFr r\r]r^r_r`rrrcrdrerfs@@r8rr/s$40V0VdrJrcHaa]tRtRtoRtV3RltV3RlRltRtVtV;t #)UVDocResNetStageimz3A ResNet stage containing multiple residual blocks.c $<\SV`4VPV,p\P!.4VnVFLwrErgVP P \TTV'd^M^V^,VVVPR74KN R#)r;)rrrrrrr@N) r1rr?r ModuleListlayersrrr@) r3config stage_indexr5rrrrr7s &&& r8rUVDocResNetStage.__init__ps| &&{3mmB' ?E ;Kx KK  " +!- *1$qL%) & 2 2 @FrJcN<V^8dQhRS[PRS[P/#rr)rFrGs"r8rHUVDocResNetStage.__annotate__#U\\ellrJc<VPF pV!V4pK V#rr)r3rlayers&& r8rUVDocResNetStage.forwards [[E!-0M!rJrrrfs@@r8rrms=$rJrcHaa]tRtRtoRtV3RltV3RlRltRtVtV;t #) UVDocResNetiz$Initial resnet_head and resnet_down.c X<\SV`4\P!.4Vn\ \ VP44FwpVPP\VPV,^,VPV,^,VP^VP^,R74Ky \P!.4Vn \ \ VP44F)p\W4pVPPV4K+ R#)r)rrr@rrN) r1rrrr>r-r+rrr@ resnet_downr?r)r3rrrr&r7s&& r8rUVDocResNet.__init__s ==,s6--./A    # # & 2 21 5a 8!'!3!3A!6q!9 & 2 2"..!3  0==, V%:%:!;r)r3rheadr&s&& r8rUVDocResNet.forwards<$$D /M%%%E!-0M&rJ)rr>rrfs@@r8rrs.+&rJrcHaa]tRtRtoRtV3RltV3RlRltRtVtV;t #)UVDocBridgeBlockizDBridge module with dilated convolutions for long-range dependencies.c <\SV`4\P!.4VnVP V,pVF+wrEVPP \WDWUR74K- R#))rrN)r1rrrblocksr*rr)r3r bridge_indexbridgerrr7s&&& r8rUVDocBridgeBlock.__init__sU mmB' %%l3%+ !K KK  ~kPXl m&,rJcd<V^8dQhRS[PRS[S[,RS[P/#r;rr4rrrrr)rFrGs"r8rHUVDocBridgeBlock.__annotate__s7||+,  rJc <VPF pV!V4pK V#rr)r3rr4blocks&&, r8rUVDocBridgeBlock.forwards" [[E!-0M!rJrrrfs@@r8r r sNnrJr cHaa]tRtRtoRtV3RltV3RlRltRtVtV;t #)UVDocPointPositions2DizDModule for predicting 2D point positions for document rectification.c <\SV`4\VP^,^,VP^,^,VP^VP^,VP VP R7Vn\P!VP^,^,VP^,^,VP^VP^,VP R7Vn R#)r)rrr@rrrrr)rrr@rrrrN) r1rrrtr@rrrqrrrconv_upr3rr7s&&r8rUVDocPointPositions2D.__init__s '44Q7:55a8;**&&!+,,(( yy44Q7:55a8;**&&!+,,   rJcN<V^8dQhRS[PRS[P/#rr)rFrGs"r8rH"UVDocPointPositions2D.__annotate__rrJcJVPV4pVPV4pV#rrr)r3rs&&r8rUVDocPointPositions2D.forwards$}5  ]3 rJr#rrfs@@r8rrsN ,rJrcZa]tRtRtoRtR]/t]P!4R4t Rt Vt R#)UVDocPreTrainedModeliTrc \P!W4\V\P4'dVP 4R#R#)zInitialize the weights.N)r _init_weights isinstancerPReLUreset_parameters)r3modules&&r8r("UVDocPreTrainedModel._init_weightss4 %%d3 fbhh ' '  # # % (rJr)N) r\r]r^r_supports_gradient_checkpointingr _can_record_outputsrno_gradr(rcrdrs@r8r&r&s3&*#) ]]_&&rJr&cXaa]tRtRtoV3Rlt]]V3RlRl44tRtVt V;t #) UVDocBridger}c<\SV`V4\P!.4Vn\ \ VP44F'pVPP\W44K) VP4R#r) r1rrrrr-r+r*rr  post_init)r3rrr7s&& r8rUVDocBridge.__init__s\  mmB' !#f&:&:";r*r num_featuresrresnetr2rr4)r3rrEr&r7s&& r8rUVDocBackbone.__init__sv  **2.r23 ))E   a ,*(!&) !&)  rJcP<V^8dQhRS[PRS[S[,RS[/#r;rr4r)r FloatTensorrrr)rFrGs"r8rHrCs5  '' +,    rJc VPV4pVP!V3/VBpRp\VP4F1wrgWpP9gKWTP V,3, pK3 \ VVP R7#)) feature_mapsrr))rFrrr.r(rr)r3rr4routputsrLr6r&s&&, r8rUVDocBackbone.forwards~ L1 ++m6v6 #D$4$45JC)))!6!6s!; == 6%!//  rJc&<V^8dQh/S[;R&#rArB)rFrGs"r8rHrCs  rJ)rrErF)r\r]r^r_has_attentionsbase_model_prefixrrr rrrbrcrdrerfs@@r8r>r>sGN"    ! 5rJr>cDaa]tRtRtoV3RltV3RlRltRtVtV;t#) UVDocHeadi)c <\SV`4\VPP4Vn\ VP^,VP ,VP^,^^^^R7Vn\V4Vn R#)r)rrr@rrrN) r1rr+rjr*num_bridge_layersrrsrrtrs&&r8rUVDocHead.__init__*sv !$V%;%;%I%I!J .//2T5K5KK003 ! &;6%B"rJcx<V^8dQhRS[PRS[S[,RS[PP/#rr)rFrGs"r8rHUVDocHead.__annotate__9s=||+,    rJc JVPV4pVPV4pV#r)rsrt)r3rr4s&&,r8rUVDocHead.forward9s* --m< 22=A rJ)rsrUrt) r\r]r^r_rrrcrdrerfs@@r8rSrS)s CrJrSz The model takes raw document images (pixel values) as input, processes them through the UVDoc backbone to predict spatial transformation parameters, and outputs the rectified (corrected) document image tensor. cdaa]tRtRtoV3RlV3Rllt]]V3RlRl44tRtVt V;t #) UVDocModeliCc <V^8dQhRS[/#rA)rh)rFrGs"r8rHUVDocModel.__annotate__Js{rJc<\SV`V4\VP4Vn\ V4VnVP4R#r)r1rr>rjr?rSr r4rs&&r8rUVDocModel.__init__Js8  %f&<&<= f%  rJc<V^8dQhRS[PRS[S[,RS[S[P,S[,/#rI)rrJrrrDr)rFrGs"r8rHr^SsF   ''  +,   u  !$B B  rJc VP!V3/VBp\P!VP^R7pVP!V3/VBp\ VVP R7#)rM)dim)r9r)r?rcatrLr rr)r3rr4backbone_outputs fused_outputsr9s&&, r8rUVDocModel.forwardQs\ ==@@ "2"?"?QG  IIm>v>-/*88  rJ)r?r ) r\r]r^r_rrrrrcrdrerfs@@r8r\r\Cs0     rJr\)r)r2r>r"rzrhr\r&)Gcollections.abcrrtorch.nnrtorch.nn.functional functionalrhuggingface_hub.dataclassesr activationsrbackbone_utilsrrrr configuration_utilsr feature_extraction_utilsr image_processing_backendsr image_transformsr r image_utilsrrmodeling_layersrmodeling_outputsrrmodeling_utilsrprocessing_utilsrutilsrrr utils.genericrrutils.import_utilsrutils.output_capturingrautorpp_lcnet.modeling_pp_lcnetr0pp_ocrv5_server_det.modeling_pp_ocrv5_server_detrr"rhrzrModulerrrr rr&r2r>rSr\__all__r)rJr8rs % .! 44;E79N-&IIC*59^;<K(-/?K(=K(\;<("(=(> :l,ll^ % :;;|ryy6"))<1(BII> &: & &I&I( % M#7%  % P 4  %   2 rJ