Fix TSR cell assignment for overlapping table boxes (#62)

* feat: TSR-aware table extraction (extract_tables_with_structure_mem)

New public function that consumes raw structure-recovery output (HTML
structure tokens + per-cell bboxes from a model like SLANet) and assembles
markdown tables by pulling cell text from the native PDF — no OCR, no
geometry inference.

Why: the existing extract_tables_in_regions_mem infers grid geometry from
text positions only and can't distinguish merged cells from multiple narrow
columns. Pairing structure recovery from a layout/TSR model with native
PDF text gets perfect text quality with proper row/col/span structure.

- New module src/tables/structured.rs: token state machine, polygon→AABB,
  crop-px→page-pt, rowspan/colspan-aware cell layout, markdown emitter.
  Accepts both 4-element rects and 8-element 4-corner polygons.
- New public extract_tables_with_structure_mem in src/lib.rs that reuses
  extract_page_text_items, region_overlaps_item, and the shared region
  text-collection helper. No existing public function modified.
- napi binding extractTablesWithStructure mirroring the existing
  extractTablesInRegions shape (f64 in JS → f32 internally).
- 14 unit tests + 5 integration tests, including a real-PDF gold-standard
  match against bits_pilani_feedback.pdf.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* TSR follow-ups: header-aware separator, cells API, v1.6.0

- cells_to_markdown emits the separator after the LAST row that contains
  is_header=true cells, falling back to "after row 0" when no header is
  flagged. Multi-row theads now render correctly. Three new unit tests
  cover: multi-row header, header not on row 0, no headers (fallback).
- New public extract_tables_with_structure_cells_mem returning
  Vec<Vec<StructuredCell>> so callers can drive their own rendering or
  debug overlays without re-doing the parse + extraction. The markdown
  variant now wraps it. The previously-unused page_pt_bbox field is
  surfaced through this API.
- New napi binding extractTablesWithStructureCells + StructuredCellJs.
- Bump @firecrawl/pdf-inspector to 1.6.0.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* fix TSR cell text assignment for overlapping bboxes

Made-with: Cursor

* bump npm package version to 1.6.1

Made-with: Cursor

---------

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
Abimael Martell
2026-04-26 16:35:19 -07:00
committed by GitHub
co-authored by Claude Opus 4.7
parent f6d5e214f1
commit 3f8fb645c9
4 changed files with 542 additions and 22 deletions
+134
View File
@@ -1484,6 +1484,82 @@ fn poly(x1: f32, y1: f32, x2: f32, y2: f32) -> Vec<f32> {
vec![x1, y1, x2, y1, x2, y2, x1, y2]
}
fn synthetic_dense_table_pdf() -> Vec<u8> {
use lopdf::content::{Content, Operation};
use lopdf::{dictionary, Document, Object, Stream};
let mut doc = Document::with_version("1.5");
let pages_id = doc.new_object_id();
let page_id = doc.new_object_id();
let font_id = doc.new_object_id();
let content_id = doc.new_object_id();
doc.objects.insert(
font_id,
dictionary! {
"Type" => "Font",
"Subtype" => "Type1",
"BaseFont" => "Helvetica",
}
.into(),
);
let operations = vec![
Operation::new("BT", vec![]),
Operation::new("Tf", vec!["F1".into(), 10.into()]),
Operation::new("Td", vec![20.into(), 700.into()]),
Operation::new("Tj", vec![Object::string_literal("Branch Name")]),
Operation::new("Td", vec![100.into(), 0.into()]),
Operation::new("Tj", vec![Object::string_literal("Deposits")]),
Operation::new("Td", vec![Object::Integer(-100), Object::Real(-16.8)]),
Operation::new("Tj", vec![Object::string_literal("Oak Street")]),
Operation::new("Td", vec![100.into(), 0.into()]),
Operation::new("Tj", vec![Object::string_literal("100")]),
Operation::new("Td", vec![Object::Integer(-100), Object::Real(-16.8)]),
Operation::new("Tj", vec![Object::string_literal("Boardwalk")]),
Operation::new("Td", vec![100.into(), 0.into()]),
Operation::new("Tj", vec![Object::string_literal("200")]),
Operation::new("ET", vec![]),
];
let content = Content { operations }.encode().unwrap();
doc.objects
.insert(content_id, Stream::new(dictionary! {}, content).into());
doc.objects.insert(
page_id,
dictionary! {
"Type" => "Page",
"Parent" => pages_id,
"MediaBox" => vec![0.into(), 0.into(), 200.into(), 800.into()],
"Resources" => dictionary! {
"Font" => dictionary! {
"F1" => font_id,
},
},
"Contents" => content_id,
}
.into(),
);
doc.objects.insert(
pages_id,
dictionary! {
"Type" => "Pages",
"Kids" => vec![page_id.into()],
"Count" => 1,
}
.into(),
);
let catalog_id = doc.add_object(dictionary! {
"Type" => "Catalog",
"Pages" => pages_id,
});
doc.trailer.set("Root", catalog_id);
let mut bytes = Vec::new();
doc.save_to(&mut bytes).unwrap();
bytes
}
#[test]
fn test_extract_tables_with_structure_real_pdf_bits_pilani() {
use pdf_inspector::{extract_tables_with_structure_mem, TsrTableInput};
@@ -1564,6 +1640,64 @@ fn test_extract_tables_with_structure_real_pdf_bits_pilani() {
);
}
#[test]
fn test_extract_tables_with_structure_dense_overlapping_slanet_boxes() {
use pdf_inspector::{extract_tables_with_structure_mem, TsrTableInput};
let buf = synthetic_dense_table_pdf();
let tokens: Vec<String> = [
"<table>",
"<thead>",
"<tr>",
"<th></th>",
"<th></th>",
"</tr>",
"</thead>",
"<tbody>",
"<tr>",
"<td></td>",
"<td></td>",
"</tr>",
"<tr>",
"<td></td>",
"<td></td>",
"</tr>",
"</tbody>",
"</table>",
]
.into_iter()
.map(String::from)
.collect();
// Rows are spaced 16.8pt apart, while the SLANet-style boxes are 40pt
// tall and overlap adjacent rows. Text must still land in only one row.
let cell_bboxes = vec![
poly(10.0, 72.0, 100.0, 112.0),
poly(90.0, 72.0, 180.0, 112.0),
poly(10.0, 88.8, 100.0, 128.8),
poly(90.0, 88.8, 180.0, 128.8),
poly(10.0, 105.6, 100.0, 145.6),
poly(90.0, 105.6, 180.0, 145.6),
];
let mds = extract_tables_with_structure_mem(
&buf,
&[TsrTableInput {
page: 0,
crop_pdf_pt_bbox: [0.0, 0.0, 200.0, 800.0],
render_dpi: 72.0,
structure_tokens: tokens,
cell_bboxes,
}],
)
.unwrap();
let expected = "|Branch Name|Deposits|\n|---|---|\n|Oak Street|100|\n|Boardwalk|200|\n";
assert_eq!(mds[0], expected);
assert!(!mds[0].contains("Branch Name Oak Street"));
assert!(!mds[0].contains("Oak Street Boardwalk"));
}
#[test]
fn test_extract_tables_with_structure_input_order_preserved() {
use pdf_inspector::{extract_tables_with_structure_mem, TsrTableInput};