docling-ibm-models 3.4.4__py3-none-any.whl → 3.5.0__py3-none-any.whl

docling_ibm_models/reading_order/reading_order_rb.py

@@ -4,15 +4,14 @@
 #
 import copy
 import logging
-import os
 import re
-from collections.abc import Iterable
 from typing import Dict, List, Set, Tuple
 
 from docling_core.types.doc.base import BoundingBox, Size
 from docling_core.types.doc.document import RefItem
 from docling_core.types.doc.labels import DocItemLabel
 from pydantic import BaseModel
+from rtree import index as rtree_index
 
 
 class PageElement(BoundingBox):
@@ -306,7 +305,13 @@ class ReadingOrderPredictor:
                     self.l2r_map[i] = j
                     self.r2l_map[j] = i
 
-    def _init_ud_maps(self, page_elems: List[PageElement]):
+    def _init_ud_maps(self, page_elems: List[PageElement]) -> None:
+        """
+        Initialize up/down maps for reading order prediction using R-tree spatial indexing.
+
+        Uses R-tree for spatial queries.
+        Determines linear reading sequence by finding preceding/following elements.
+        """
         self.up_map = {}
         self.dn_map = {}
 
@@ -314,51 +319,83 @@ class ReadingOrderPredictor:
             self.up_map[i] = []
             self.dn_map[i] = []
 
-        for j, pelem_j in enumerate(page_elems):
+        # Build R-tree spatial index
+        spatial_idx = rtree_index.Index()
+        for i, pelem in enumerate(page_elems):
+            spatial_idx.insert(i, (pelem.l, pelem.b, pelem.r, pelem.t))
 
+        for j, pelem_j in enumerate(page_elems):
             if j in self.r2l_map:
                 i = self.r2l_map[j]
-
                 self.dn_map[i] = [j]
                 self.up_map[j] = [i]
-
                 continue
 
-            for i, pelem_i in enumerate(page_elems):
+            # Find elements above current that might precede it in reading order
+            query_bbox = (pelem_j.l - 0.1, pelem_j.t, pelem_j.r + 0.1, float("inf"))
+            candidates = list(spatial_idx.intersection(query_bbox))
 
+            for i in candidates:
                 if i == j:
                     continue
 
-                is_horizontally_connected: bool = False
-                is_i_just_above_j: bool = pelem_i.overlaps_horizontally(
-                    pelem_j
-                ) and pelem_i.is_strictly_above(pelem_j)
-
-                for w, pelem_w in enumerate(page_elems):
-
-                    if not is_horizontally_connected:
-                        is_horizontally_connected = pelem_w.is_horizontally_connected(
-                            pelem_i, pelem_j
-                        )
+                pelem_i = page_elems[i]
 
-                    # ensure there is no other element that is between i and j vertically
-                    if is_i_just_above_j and (
-                        pelem_i.overlaps_horizontally(pelem_w)
-                        or pelem_j.overlaps_horizontally(pelem_w)
-                    ):
-                        i_above_w: bool = pelem_i.is_strictly_above(pelem_w)
-                        w_above_j: bool = pelem_w.is_strictly_above(pelem_j)
-
-                        is_i_just_above_j = not (i_above_w and w_above_j)
-
-                if is_i_just_above_j:
+                # Check spatial relationship
+                if not (
+                    pelem_i.is_strictly_above(pelem_j)
+                    and pelem_i.overlaps_horizontally(pelem_j)
+                ):
+                    continue
 
+                # Check for interrupting elements
+                if not self._has_sequence_interruption(
+                    spatial_idx, page_elems, i, j, pelem_i, pelem_j
+                ):
+                    # Follow left-to-right mapping
                     while i in self.l2r_map:
                         i = self.l2r_map[i]
 
                     self.dn_map[i].append(j)
                     self.up_map[j].append(i)
 
+    def _has_sequence_interruption(
+        self,
+        spatial_idx: rtree_index.Index,
+        page_elems: List[PageElement],
+        i: int,
+        j: int,
+        pelem_i: PageElement,
+        pelem_j: PageElement,
+    ) -> bool:
+        """Check if elements interrupt the reading sequence between i and j."""
+        # Query R-tree for elements between i and j
+        x_min = min(pelem_i.l, pelem_j.l) - 1.0
+        x_max = max(pelem_i.r, pelem_j.r) + 1.0
+        y_min = pelem_j.t
+        y_max = pelem_i.b
+
+        candidates = list(spatial_idx.intersection((x_min, y_min, x_max, y_max)))
+
+        for w in candidates:
+            if w in (i, j):
+                continue
+
+            pelem_w = page_elems[w]
+
+            # Check if w interrupts the i->j sequence
+            if (
+                (
+                    pelem_i.overlaps_horizontally(pelem_w)
+                    or pelem_j.overlaps_horizontally(pelem_w)
+                )
+                and pelem_i.is_strictly_above(pelem_w)
+                and pelem_w.is_strictly_above(pelem_j)
+            ):
+                return True
+
+        return False
+
     def _do_horizontal_dilation(self, page_elems, dilated_page_elems):
 
         for i, pelem_i in enumerate(dilated_page_elems):

docling_ibm_models/tableformer/models/table04_rs/transformer_rs.py

@@ -2,6 +2,8 @@
 # Copyright IBM Corp. 2024 - 2024
 # SPDX-License-Identifier: MIT
 #
+
+
 import logging
 import math
 from typing import Optional
@@ -99,6 +101,7 @@ class TMTransformerDecoderLayer(nn.TransformerDecoderLayer):
             tgt,
             attn_mask=None,  # None, because we only care about the last tag
             key_padding_mask=tgt_key_padding_mask,
+            need_weights=False,  # Optimization: Don't compute attention weights
         )[0]
         tgt_last_tok = tgt_last_tok + self.dropout1(tmp_tgt)
         tgt_last_tok = self.norm1(tgt_last_tok)
@@ -110,6 +113,7 @@ class TMTransformerDecoderLayer(nn.TransformerDecoderLayer):
                 memory,
                 attn_mask=memory_mask,
                 key_padding_mask=memory_key_padding_mask,
+                need_weights=False,  # Optimization: Don't compute attention weights
             )[0]
             tgt_last_tok = tgt_last_tok + self.dropout2(tmp_tgt)
             tgt_last_tok = self.norm2(tgt_last_tok)

{docling_ibm_models-3.4.4.dist-info → docling_ibm_models-3.5.0.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: docling-ibm-models
-Version: 3.4.4
+Version: 3.5.0
 Summary: This package contains the AI models used by the Docling PDF conversion package
 Author-email: Nikos Livathinos <nli@zurich.ibm.com>, Maxim Lysak <mly@zurich.ibm.com>, Ahmed Nassar <ahn@zurich.ibm.com>, Christoph Auer <cau@zurich.ibm.com>, Michele Dolfi <dol@zurich.ibm.com>, Peter Staar <taa@zurich.ibm.com>
 License-Expression: MIT
@@ -11,11 +11,17 @@ Project-URL: changelog, https://github.com/docling-project/docling-ibm-models/bl
 Keywords: docling,convert,document,pdf,layout model,segmentation,table structure,table former
 Classifier: Operating System :: MacOS :: MacOS X
 Classifier: Operating System :: POSIX :: Linux
+Classifier: Operating System :: Microsoft :: Windows
 Classifier: Development Status :: 5 - Production/Stable
 Classifier: Intended Audience :: Developers
 Classifier: Intended Audience :: Science/Research
 Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
 Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.9
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
 Requires-Python: <4.0,>=3.9
 Description-Content-Type: text/markdown
 License-File: LICENSE
@@ -31,6 +37,7 @@ Requires-Dist: pydantic<3.0.0,>=2.0.0
 Requires-Dist: docling-core<3.0.0,>=2.19.0
 Requires-Dist: transformers<5.0.0,>=4.42.0
 Requires-Dist: numpy<3.0.0,>=1.24.4
+Requires-Dist: rtree>=1.0.0
 Dynamic: license-file
 
 [![PyPI version](https://img.shields.io/pypi/v/docling-ibm-models)](https://pypi.org/project/docling-ibm-models/)

{docling_ibm_models-3.4.4.dist-info → docling_ibm_models-3.5.0.dist-info}/RECORD

@@ -11,7 +11,7 @@ docling_ibm_models/document_figure_classifier_model/document_figure_classifier_p
 docling_ibm_models/layoutmodel/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 docling_ibm_models/layoutmodel/layout_predictor.py,sha256=ArVgs7FBOiu23TC-JoybcaTp7F7a4BgYC8uRVxTgx4E,5681
 docling_ibm_models/reading_order/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-docling_ibm_models/reading_order/reading_order_rb.py,sha256=Vk3ufc47w2FnVaLI5UGpxoBTZFcpWuIrSAaNGa9c5Rg,20416
+docling_ibm_models/reading_order/reading_order_rb.py,sha256=RpcR0Q1oeF3JK-j6O0KyNZtGgBeqKUHsIOj7hmPumUo,21670
 docling_ibm_models/tableformer/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 docling_ibm_models/tableformer/common.py,sha256=2zgGZBFf4fXytEaXrZR2NU6FWdX2kxO0DHlGZmuvpNQ,3230
 docling_ibm_models/tableformer/otsl.py,sha256=DxEwJVC_IqomZs_wUzj-TWjUUgQuEVcm8MXru7VYGkA,21391
@@ -29,13 +29,13 @@ docling_ibm_models/tableformer/models/table04_rs/__init__.py,sha256=47DEQpj8HBSa
 docling_ibm_models/tableformer/models/table04_rs/bbox_decoder_rs.py,sha256=JV9rFh9caT3qnwWlZ0CZpw5aiiNzyTbfVp6H6JMxS0Q,6117
 docling_ibm_models/tableformer/models/table04_rs/encoder04_rs.py,sha256=iExmqJ0Pn0lJU3nWb_x8abTn42GctMqE55_YA2ppgvc,1975
 docling_ibm_models/tableformer/models/table04_rs/tablemodel04_rs.py,sha256=Mv17JGgO12hIt8jrnflWLgOimdFYkBLuV0rxaGawBpk,12266
-docling_ibm_models/tableformer/models/table04_rs/transformer_rs.py,sha256=2i5qqVijyic2VeMI0d1-9gVg2vTbxfi9Ciyo-r41iOY,6464
+docling_ibm_models/tableformer/models/table04_rs/transformer_rs.py,sha256=XW2k43MjwjrgrPaukuwemX2k03dyGpy1YvRpkKYvkAY,6632
 docling_ibm_models/tableformer/utils/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 docling_ibm_models/tableformer/utils/app_profiler.py,sha256=Pb7o1zcikKXh7ninaNt4_nVa1xuUrogZxbTr6U6jkEE,8392
 docling_ibm_models/tableformer/utils/mem_monitor.py,sha256=NFZUnrfLThXNZQrm3ESRmPSJmPF2J1z3E2v_72O4dRw,6408
 docling_ibm_models/tableformer/utils/utils.py,sha256=8Bxf1rEn977lFbY9NX0r5xh9PvxIRipQZX_EZW92XfA,10980
-docling_ibm_models-3.4.4.dist-info/licenses/LICENSE,sha256=mBb7ErEcM8VS9OhiGHnQ2kk75HwPhr54W1Oiz3965MY,1088
-docling_ibm_models-3.4.4.dist-info/METADATA,sha256=_gfa7Fz_yuyE1m_t06O04YuroSoSAUxlvVRO3WWTBOg,6370
-docling_ibm_models-3.4.4.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-docling_ibm_models-3.4.4.dist-info/top_level.txt,sha256=tIB9D3naeP7s92RAs1d9SPaHc4S4iQIepjtbkf5Q5g0,19
-docling_ibm_models-3.4.4.dist-info/RECORD,,
+docling_ibm_models-3.5.0.dist-info/licenses/LICENSE,sha256=mBb7ErEcM8VS9OhiGHnQ2kk75HwPhr54W1Oiz3965MY,1088
+docling_ibm_models-3.5.0.dist-info/METADATA,sha256=fyDgSAuTt6vIhYa4HdFZIDrEMPMzmMJp5QhqWCAhO6E,6705
+docling_ibm_models-3.5.0.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+docling_ibm_models-3.5.0.dist-info/top_level.txt,sha256=tIB9D3naeP7s92RAs1d9SPaHc4S4iQIepjtbkf5Q5g0,19
+docling_ibm_models-3.5.0.dist-info/RECORD,,