jMSSKrSSKrSSKJr SSKJr SSKJr SSKJ r SSK r SSK J r JrJr SSKJrJrJrJrJrJr SSKJrJr SS KJrJrJr SS KJr SS K J!r! SS K"J#r# SS K$J%r%J&r& SSK'J(r( SSK)J*r* SSK+J,r,J-r- \R\"\/5r0S\!S\4Sjr1SSK2r2Sr3"SS\5\5r6\7"\Rp\Rr15r:Sr;S\\S\<\=\\=\44Sjr>"SS\-\,5r?g!\4a Sr3NZf=f)N)abstractmethod)Iterable)Enum)Path) BoundingBox CoordOriginSize)BoundingRectanglePdfCellRenderingModePdfPageGeometry PdfTextCellSegmentedPdfPageTextCell)Image ImageDraw)binary_dilation find_objectslabel)AcceleratorOptions)Page)ConversionResult)OcrMode OcrOptions)settings)BoundingBoxSpatialIndex)BaseModelWithOptions BasePageModelpagereturnc URRURRp![SSUSUUSU[R S9 n[ SX!S[R S9n[[SUSUUUUUS9///S9$)zA minimal SegmentedPdfPage for pages whose native parse was skipped (PagePreprocessingOptions.skip_cell_extraction), sized from the page.r) r_x0r_y0r_x1r_y1r_x2r_y2r_x3r_y3 coord_originltrbr)gcrop_box)anglerect boundary_typeart_bbox bleed_bbox crop_bbox media_bbox trim_bbox) dimension char_cells word_cellstextline_cells) sizewidthheightr r BOTTOMLEFTrrr )rr=r>r1bboxs Z/home/mande/repo/quber/.venv/lib/python3.13/site-packages/docling/models/base_ocr_model.py_empty_segmented_pagerB#sIIOOTYY%5%56          ++ D fAWAW XD !$   TFc\rSrSrSrSrSrg)_MergeCellsPriorityVpdf_cells_firstocr_cells_firstN)__name__ __module__ __qualname____firstlineno__ PDF_FIRST OCR_FIRST__static_attributes__rIrCrArErEVs!I"IrCrEcellsc/n/nUHPn[U[5(a'UR[;aUR U5 M?UR U5 MR X4$)zSplit cells into (visible, invisible), preserving the order within each group. Cells carrying no rendering mode (plain `TextCell`, e.g. OCR output) count as visible, as does `UNKNOWN` (-1): no `Tr` operator means the PDF default mode 0 (fill). ) isinstancer rendering_mode_INVISIBLE_RENDERING_MODESappend)rQvisible invisiblecells rA_segregate_by_visibilityrZhsY!G "I t[ ) )##'AA   T " NN4   rCc \rSrSrSrSrS\S\S-S\S\ 4S jr S \ S \ \ 4S jrS \ S \ \ 4S jrS \ S \ \ 4SjrS"S\S\\ S \\\ \ 44SjjrS#S\ \S \ S\S\S-S S4 SjjrS$S\ \S\ \S\S\S \ \4 SjjrS\ \S\ \S \ \4SjrS%S\4Sjjr\S\S\\ S \\ 4Sj5r\\S \ \4S j55r!S!r"g)& BaseOcrModel}zE Base class for all OCR models. It offers common OCR functionalities enabledartifacts_pathNoptionsaccelerator_optionscXlX0lgNr_ra)selfr_r`rarbs rA__init__BaseOcrModel.__init__s  rCrrc8URceURR[R:Xd(URR[R :XaUR U5nU$URR[R:XaURU5nU$URR[R:XaD[SSURRURR[RS9/nW$)zM Produce the input rects for the OCR according to the logic for each OcrMode rr*)r<ramoderDEFAULTPDF_AWARE_LAYOUT_REGIONS _find_pdf_aware_layout_ocr_rectsLAYOUT_REGIONS_find_layout_ocr_rects FULL_PAGErr=r>rTOPLEFT)rfr ocr_rectss rA get_ocr_rectsBaseOcrModel.get_ocr_rectssyy$$$ LL   0||  G$D$DD==dCI\\  '"8"8 833D9I\\  '"3"3 3iiooii&&!,!4!4 IrCc URRc/$URRRVs/sHo"RPM nnUR UR U[ RS9upCU$s snf)zW 1. Collect the bboxes of all layout clusters. 2. Deduplicate the candidate ocr_rects.  dilation_size) predictionslayoutclustersr@_deduplicate_rectsr<r\DEFAULT_DILATION_SIZE)rfrcrr_s rAro#BaseOcrModel._find_layout_ocr_rectss    " " *I&*%5%5%<%<%E%EF%EVV%E F.. IIy 0R0R/  GsBc>URRc/$URcURU5$URceURn[ SSURR URR[RS9nURUS9SLnSnSnU(dUR5nUcUR5n[5n[U5H.upURXR R#55 M0 [%UR'55n UR)5n U bU R+U 5 [5n[U 5HupURX5 M /n URRR,HnUR.nU(aURUSSSS9nO&Uce[1SUR3U555nU(aU R5U5 MiU(aURUSSSS9nO&Uce[1S UR3U555nU(aMU R5U5 M UR7URU [8R:S 9unn U $) a Compute the OCR rects from the layout clusters of a programmatic PDF. 1. Start from the layout clusters. 2. Eliminate clusters that intersect exclusively with programmatic text PDF cells The following clusters therefore remain: - Clusters without any overlapping PDF cell. - Clusters with at least one overlapping non-text region (e.g. bitmap, shape). 3. Deduplicate the remaining cluster bboxes. Nrr*)r@FT)r@charsshapesbitmapsc3&# UHnSv M g7fTNrI.0r~s rA @BaseOcrModel._find_pdf_aware_layout_ocr_rects..s#"KQD"Kc3&# UHnSv M g7frrIrs rArr sS-Rt-Rrrv)rxry_backendror<rr=r>rrqhas_content_inget_visible_text_cellsget_text_cellsr enumerateinsertr1to_bounding_boxlistget_bitmap_rects"get_connected_shape_bounding_boxesextendrzr@any intersectionrVr{r\r|)rfrbackend page_bboxuse_backend_queries text_indexnon_text_index text_cellsi text_cellnon_text_boxes shape_boxesr@rrcluster cluster_bbox has_non_texthas_textr~s rArm-BaseOcrModel._find_pdf_aware_layout_ocr_rectss    " " *I == ..t4 4yy$$$-- iiooii$,,  &44)4DDP59 9=" 779J!$335 02J )* 5 !!!^^%C%C%EF!6"'":":"<=N!DDFK&%%k246N$^4%%a.5(* ''..77G"<rDrawas_tuple rectanglenparrayonesuint8 CV2_INSTALLEDcv2dilateastyperrrrstartstoprrqsum)rfr<rrwimagedrawr1x0y0x1y1np_imagekernel labeled_imager~slicesslcbounding_boxes area_fracs rAr{BaseOcrModel._deduplicate_rectss % #U4;;%78  ~~e$D!]]_NBB"2Yb 59eBiGNBB NNRHrh/aN 8 88E? 1 WWm;288LF}::\))"((3V+8a<6J!A. m,   a&,,a&,,a&++/a&++/(00     FF8a<(DJJ,DE >**  s!AG9 ocr_cellsconv_resprioritycDURnURR[R:XaUnOaUcHURR[R :Xa[ RO[ RnURUUU[S9n[U5H upxXxl M URc[U5Ul XbRl[!U5S:URlURR[R:XaURR$V s/sHoR&(dMU PM sn URlURR(V s/sHoR&(dMU PM sn URl[!URR$5S:URl[!URR(5S:URlUV s/sH!oR&(dMU R.PM# n n U (aF[1[2R4"U 55UR.R6UR8lggs sn fs sn fs sn f)a Post-process the OCR cells and update the page object according to the algorithm: - If FULL_PAGE: Any existing PDF cells are ignored and only the OCR cells are used. - If LAYOUT_REGIONS or PDF_AWARE_LAYOUT_REGIONS and the priority parameter is None, the priority is auto-selected based on the OcrMode: - OCR_FIRST when LAYOUT_REGIONS - PDF_FIRST when PDF_AWARE_LAYOUT_REGIONS Check the comments on _MergeCellsPriority for the semantic of each priority value N)segregate_pdf_cellsr)rQrarjrrprnrErOrN_merge_ocr_and_pdf_cells&_PDF_CELLS_POST_PROCESSING_SEGREGATIONrindex parsed_pagerBr;len has_linesr:from_ocrr9 has_words has_chars confidencefloatrmeanpagespage_no ocr_score) rfrrrrexisting_cells final_cellsrrYr}ocr_confidencess rApost_process_cellsBaseOcrModel.post_process_cellsPs$ <<   1 1 1#K||((G,B,BB(11,66 77$J 8K!-GAJ.    # 5T:D +6'%(%5%9" <<   1 1 1++66+6a**6+D   ' ++66+6a**6+D   '*-T-=-=-H-H)IA)MD   &),T-=-=-H-H)IA)MD   &1<KA <1<<K @E(AH   % %dll 3 = ++ Ls$=JJJJJ3J pdf_cellsrcU(a[U5upVO [U5/peU[R:XaXQpOXpUR Xx5n U(aUR X5n U $)a! Merge PDF and OCR cells, resolving overlaps according to `priority`. When `segregate_pdf_cells` is True, the OCR cells are merged only with the visible PDF cells and the invisible ones are put back afterwards. Otherwise, the OCR cells are merged with all the PDF cells indiscriminately. )rZrrErN_merge_cells_by_priority) rfrrrr visible_cellsinvisible_cellsprioritized_cellssecondary_cells merged_cellss rAr%BaseOcrModel._merge_ocr_and_pdf_cellssh -Ei-P *M?-1)_b? *44 41>1:445FX  88WLrCrrc^[5n[U5n[U5[U5::a[U5H.upVUR XVR R 55 M0 UHVn[SURUR R 5555nU(aMEURU5 MX U$[U5H.upVUR XVR R 55 M0 [5mUH;nTRURUR R 555 M= URU4Sj[U555 U$)zR Keep every prioritized cell, plus the secondary cells that overlap none of them. c3&# UHnSv M g7frrIrs rAr8BaseOcrModel._merge_cells_by_priority..s"OQD"Orc3<># UHupUT;dM Uv M g7frdrI)rrrYoverlapping_idss rArrs$ 9GAO+9s  ) rrrrrr1rrrrVsetupdater) rfrridxrrrYoverlapsrs @rAr%BaseOcrModel._merge_cells_by_prioritys;&'-.  !S%9 9%%67 1ii779:8'"%"2"24993L3L3N"O x ''- ((%_5 1ii779:6(+O)&&s'7'7 8Q8Q8S'TU*    (9   rCshowc[R"UR5nURURR- nUR URR - n[ R"US5nUH=n U R5uppX-n X-n X-n X-n SnURX4X4/USS9 M? URHnnURR5R5uppX-n X-n X-n X-n X::aXpUR(aSOSnURX4X4/US9 Mp U(aUR5 g[[ R"R$5SUR&R(R*3- nUR-S S S 9 US UR.S S 3- nUR1[3U5SS9 g)z] - OCR input rects: Yellow panes - OCR detected text: Magenta bboxes - PDF text: Gray bboxes RGBA)rr(N)routlinemagentagray)rdebug_T)parentsexist_ok ocr_page_05z.pngpng)format)copydeepcopyrr=r<r>rrrrrQr1rrrrrdebugdebug_output_pathinputfilestemmkdirrsavestr)rfrrrrrrscale_xscale_yrr1rrrr shade_colortccolorout_pathout_files rAdraw_ocr_rects_and_cells%BaseOcrModel.draw_ocr_rects_and_cellss  djj)++ /,,!1!11~~eV,D!]]_NBB MB MB MB MB+K NNRHrh/k4N P**BWW446??ANBB MB MB MB MBxB!#I&E NNRHrh/N ?  JJLX^^5568>>..33456  NN4$N 7Idll2->d"CCH JJs8}UJ 3rC page_batchcgrdrI)rfrrs rA__call__BaseOcrModel.__call__s rCcgrdrI)clss rAget_options_typeBaseOcrModel.get_options_types rCre)rrd)T)F)#rJrKrLrM__doc__r|boolrrrrgrrrrsrormr rtuplerr{rrrErrrrrr classmethodtyper"rPrIrCrAr\r\}s   t    0 $4 +<<4D4E"WTWd;>OWtGH6+6+!)+!66+ ud;'' (6+z04 A>AA# A &, A  AP%) ">">"& " " " h "H&>&h& h &P0404d ( 6>tn $   j!1  rCr\)@r loggingabcrcollections.abcrenumrpathlibrnumpyrdocling_core.types.docrrr docling_core.types.doc.pager r r r rrPILrr scipy.ndimagerrr%docling.datamodel.accelerator_optionsrdocling.datamodel.base_modelsrdocling.datamodel.documentr"docling.datamodel.pipeline_optionsrrdocling.datamodel.settingsrdocling.datamodel.spatialrdocling.models.base_modelrr getLoggerrJ_logrBrr ImportErrorrrE frozenset INVISIBLE ONLY_CLIPPINGrUrr&rrZr\rIrCrAr@s  $AA!>>D.7B/=I"(()9(VM "#t"'##%9%G%GH *.& H  4>4> )**] ="6] WMsC11C<;C<