fix: Empty table handling (#2365)

* add table raw cells when no table structure model was used

Signed-off-by: Michele Dolfi <[email protected]>

* Add RichTableCell instance for tables with missing structure.

Signed-off-by: Christoph Auer <[email protected]>

* Update test GT

Signed-off-by: Christoph Auer <[email protected]>

* update test results

Signed-off-by: Michele Dolfi <[email protected]>

---------

Signed-off-by: Michele Dolfi <[email protected]>
Signed-off-by: Christoph Auer <[email protected]>
Co-authored-by: Michele Dolfi <[email protected]>

Author: cau-git

docling/models/readingorder_model.py

@@ -216,8 +216,13 @@ def _readingorder_elements_to_docling_doc(
             elif isinstance(element, Table):
                 # Check if table has no structure prediction
                 if element.num_rows == 0 and element.num_cols == 0:
-                    # Create minimal 1x1 table with rich cell containing all children
-                    tbl_data = TableData(num_rows=1, num_cols=1, table_cells=[])
+                    # Only create 1x1 table if there are children to put in it
+                    if element.cluster.children:
+                        # Create minimal 1x1 table with rich cell containing all children
+                        tbl_data = TableData(num_rows=1, num_cols=1, table_cells=[])
+                    else:
+                        # Create empty table with no structure
+                        tbl_data = TableData(num_rows=0, num_cols=0, table_cells=[])
                 else:
                     tbl_data = TableData(
                         num_rows=element.num_rows,
@@ -253,8 +258,12 @@ def _readingorder_elements_to_docling_doc(
 
                         tbl.footnotes.append(new_footnote_item.get_ref())
 
-                # Handle case where table has no structure prediction
-                if element.num_rows == 0 and element.num_cols == 0:
+                # Handle case where table has no structure prediction but has children
+                if (
+                    element.num_rows == 0
+                    and element.num_cols == 0
+                    and element.cluster.children
+                ):
                     # Create rich cell containing all child elements
                     rich_cell_ref = self._create_rich_cell_group(element, out_doc, tbl)
 

tests/data/groundtruth/docling_v2/2203.01017v2.doctags.txt

@@ -218,7 +218,7 @@
 <picture><loc_67><loc_92><loc_205><loc_308></picture>
 <caption><loc_52><loc_317><loc_223><loc_323>Figure 8: Example of a table with multi-line header.</caption>
 <picture><loc_252><loc_59><loc_455><loc_185><caption><loc_252><loc_194><loc_445><loc_207>Figure 9: Example of a table with big empty distance between cells.</caption></picture>
-<otsl><loc_274><loc_286><loc_400><loc_317><ecel><nl></otsl>
+<otsl><loc_274><loc_286><loc_400><loc_317></otsl>
 <picture><loc_273><loc_239><loc_424><loc_420><caption><loc_255><loc_430><loc_443><loc_435>Figure 10: Example of a complex table with empty cells.</caption></picture>
 <page_footer><loc_239><loc_464><loc_247><loc_469>13</page_footer>
 <page_break>

tests/data/groundtruth/docling_v2/2203.01017v2.json

@@ -829,16 +829,6 @@
       "content_layer": "body",
       "name": "list",
       "label": "list"
-    },
-    {
-      "self_ref": "#/groups/12",
-      "parent": {
-        "$ref": "#/tables/6"
-      },
-      "children": [],
-      "content_layer": "body",
-      "name": "rich_cell_group_7_0_0",
-      "label": "unspecified"
     }
   ],
   "texts": [
@@ -25418,11 +25408,7 @@
       "parent": {
         "$ref": "#/body"
       },
-      "children": [
-        {
-          "$ref": "#/groups/12"
-        }
-      ],
+      "children": [],
       "content_layer": "body",
       "label": "table",
       "prov": [
@@ -25445,43 +25431,10 @@
       "references": [],
       "footnotes": [],
       "data": {
-        "table_cells": [
-          {
-            "row_span": 1,
-            "col_span": 1,
-            "start_row_offset_idx": 0,
-            "end_row_offset_idx": 1,
-            "start_col_offset_idx": 0,
-            "end_col_offset_idx": 1,
-            "text": "",
-            "column_header": false,
-            "row_header": false,
-            "row_section": false,
-            "fillable": false,
-            "ref": {
-              "$ref": "#/groups/12"
-            }
-          }
-        ],
-        "num_rows": 1,
-        "num_cols": 1,
-        "grid": [
-          [
-            {
-              "row_span": 1,
-              "col_span": 1,
-              "start_row_offset_idx": 0,
-              "end_row_offset_idx": 1,
-              "start_col_offset_idx": 0,
-              "end_col_offset_idx": 1,
-              "text": "",
-              "column_header": false,
-              "row_header": false,
-              "row_section": false,
-              "fillable": false
-            }
-          ]
-        ]
+        "table_cells": [],
+        "num_rows": 0,
+        "num_cols": 0,
+        "grid": []
       },
       "annotations": []
     }

tests/data/groundtruth/docling_v2/2203.01017v2.md

@@ -367,9 +367,6 @@ Figure 9: Example of a table with big empty distance between cells.
 
 <!-- image -->
 
-|    |
-|----|
-
 Figure 10: Example of a complex table with empty cells.
 
 <!-- image -->