Datasets:
File size: 3,789 Bytes
eb58dca | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 | # /// script
# requires-python = ">=3.10"
# dependencies = []
# ///
"""Export VietFinTab to an image -> HTML dataset (plain <table>, no CSS).
Structure comes from the XML (rows x columns grid, spanning cells -> rowspan/colspan);
text comes from the content JSON, so only the tables with a content label are exported.
Usage: uv run scripts/export_html.py [--root .] [--out datasets/VietFinTab-image2html]
"""
import argparse
import json
import shutil
from html import escape
from pathlib import Path
from export_yolo import boxes
def inside(b, x, y):
return b[0] <= x <= b[2] and b[1] <= y <= b[3]
def cell_text(tokens):
"""tokens: [(text, (x0, y0, x1, y1))] -> text in reading order (lines by y, then x)."""
lines = []
for t, b in sorted(tokens, key=lambda t: t[1][1] + t[1][3]):
cy, h = (b[1] + b[3]) / 2, b[3] - b[1]
if lines and cy - lines[-1][0] <= h / 2:
lines[-1][1].append((b[0], t))
else:
lines.append((cy, [(b[0], t)]))
return " ".join(t for _, line in lines for _, t in sorted(line))
def build_html(rows, cols, spans, ocr):
rows, cols = sorted(rows, key=lambda b: b[1]), sorted(cols, key=lambda b: b[0])
center = lambda i, j: ((cols[j][0] + cols[j][2]) / 2, (rows[i][1] + rows[i][3]) / 2)
anchor, covered = {}, set() # anchor (i, j) -> (rowspan, colspan, key); covered = merged-away cells
owner = {} # (i, j) -> anchor of the cell that contains it
for s in spans:
grid = [(i, j) for i in range(len(rows)) for j in range(len(cols)) if inside(s, *center(i, j))]
if not grid:
continue
i0, j0 = min(i for i, _ in grid), min(j for _, j in grid)
anchor[(i0, j0)] = (max(i for i, _ in grid) - i0 + 1, max(j for _, j in grid) - j0 + 1)
covered |= set(grid) - {(i0, j0)}
owner.update({g: (i0, j0) for g in grid})
tokens = {}
for t in ocr:
b = t["bbox"]
cx, cy = (b[0] + b[2]) / 2, (b[1] + b[3]) / 2
i = next((i for i, r in enumerate(rows) if r[1] <= cy <= r[3]), None)
j = next((j for j, c in enumerate(cols) if c[0] <= cx <= c[2]), None)
if i is not None and j is not None: # tokens outside the grid are dropped
tokens.setdefault(owner.get((i, j), (i, j)), []).append((t["text"], b))
out = ["<!DOCTYPE html>", '<meta charset="utf-8">', '<table border="1">']
for i in range(len(rows)):
tds = []
for j in range(len(cols)):
if (i, j) in covered:
continue
rs, cs = anchor.get((i, j), (1, 1))
attrs = (f' rowspan="{rs}"' if rs > 1 else "") + (f' colspan="{cs}"' if cs > 1 else "")
tds.append(f"<td{attrs}>{escape(cell_text(tokens.get((i, j), [])))}</td>")
out.append(f"<tr>{''.join(tds)}</tr>")
return "\n".join(out + ["</table>", ""])
def export(root, out):
meta = [json.loads(l) for l in (root / "metadata.jsonl").read_text().splitlines()]
meta = [m for m in meta if m["has_content_label"]]
for d in ("images", "html"):
(out / d).mkdir(parents=True, exist_ok=True)
for m in meta:
_, _, rows, cols, spans, _ = boxes(root / m["structure_file"])
ocr = json.loads((root / m["content_file"]).read_text())["ocr"]
shutil.copy2(root / m["file_name"], out / "images" / f"{m['id']}.jpg")
(out / "html" / f"{m['id']}.html").write_text(build_html(rows, cols, spans, ocr), encoding="utf-8")
print(f"{len(meta)} tables -> {out}/")
if __name__ == "__main__":
ap = argparse.ArgumentParser()
ap.add_argument("--root", type=Path, default=Path(__file__).resolve().parent.parent)
ap.add_argument("--out", type=Path, default=Path("datasets/VietFinTab-image2html"))
a = ap.parse_args()
export(a.root, a.out)
|