Spaces:
Paused
Paused
File size: 13,446 Bytes
f66643d | 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 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 | """Phase-3 render smoke + feasibility test.
Runs in two modes:
1. Pre-flight (default) — validates the plan's assumptions against the real PDF
+ parsed JSON without needing pdf2zh.render to exist yet:
* inputs load, schema is correct
* PyMuPDF + numpy can access pixmap samples
* detect_bg_color / detect_text_color produce sensible values on a real bbox
* font registration + text insertion work on a synthetic page
* stats: overflow risk, native-text presence, redaction necessity
2. Render (--render --font ... --output ...) — once pdf2zh.render is built,
runs the full pipeline and verifies the output PDF is non-empty and the
redacted+rendered text is extractable.
Usage:
python test/verify_render.py \
--input GK_ChuanMucKeToan_Nhom02.pdf \
--parsed output-3.translated.json
python test/verify_render.py \
--input GK_ChuanMucKeToan_Nhom02.pdf \
--parsed output-3.translated.json \
--render --font fonts/NotoSans-Regular.ttf --output GK.rendered.pdf
"""
from __future__ import annotations
import argparse
import importlib
import json
import re
import sys
from collections import Counter
from pathlib import Path
# Ensure project root is on sys.path when script is run directly.
sys.path.insert(0, str(Path(__file__).parent.parent))
import fitz
import numpy as np
# ---------------------------------------------------------------------------
# Pure helpers — duplicated here so the script runs without pdf2zh.render.
# Once color.py exists, the renderer should produce identical results.
# ---------------------------------------------------------------------------
def _pixmap_to_array(pm: fitz.Pixmap) -> np.ndarray:
arr = np.frombuffer(pm.samples, dtype=np.uint8).reshape(pm.height, pm.width, pm.n)
if pm.n == 4:
arr = arr[:, :, :3]
return arr
def _bbox_to_pixels(bbox, pw, ph, pm):
sx = pm.width / pw
sy = pm.height / ph
x0, y0, x1, y1 = bbox
px0 = max(0, int(round(x0 * sx)))
py0 = max(0, int(round(y0 * sy)))
px1 = min(pm.width, int(round(x1 * sx)))
py1 = min(pm.height, int(round(y1 * sy)))
return px0, py0, px1, py1
def detect_bg_color(arr, bbox_px, edge=2, qstep=16):
px0, py0, px1, py1 = bbox_px
if px1 - px0 < 2 * edge + 1 or py1 - py0 < 2 * edge + 1:
return (255, 255, 255)
top = arr[py0 : py0 + edge, px0:px1]
bot = arr[py1 - edge : py1, px0:px1]
left = arr[py0 + edge : py1 - edge, px0 : px0 + edge]
right = arr[py0 + edge : py1 - edge, px1 - edge : px1]
band = np.concatenate(
[
top.reshape(-1, 3),
bot.reshape(-1, 3),
left.reshape(-1, 3),
right.reshape(-1, 3),
]
)
if band.size == 0:
return (255, 255, 255)
q = (band // qstep) * qstep + qstep // 2
keys = q[:, 0].astype(np.int32) * 65536 + q[:, 1].astype(np.int32) * 256 + q[:, 2]
vals, counts = np.unique(keys, return_counts=True)
winner = vals[counts.argmax()]
return (int((winner >> 16) & 0xFF), int((winner >> 8) & 0xFF), int(winner & 0xFF))
def detect_text_color(arr, bbox_px, bg, edge=2, qstep=16, dist=32, min_ratio=0.05):
px0, py0, px1, py1 = bbox_px
inner = arr[py0 + edge : py1 - edge, px0 + edge : px1 - edge]
if inner.size == 0:
return (0, 0, 0)
flat = inner.reshape(-1, 3).astype(np.int32)
bg_arr = np.array(bg, dtype=np.int32)
d = np.sqrt(((flat - bg_arr) ** 2).sum(axis=1))
keep = flat[d > dist]
if keep.size == 0:
return (0, 0, 0)
q = (keep // qstep) * qstep + qstep // 2
keys = q[:, 0] * 65536 + q[:, 1] * 256 + q[:, 2]
vals, counts = np.unique(keys, return_counts=True)
idx = counts.argmax()
if counts[idx] / max(1, len(flat)) < min_ratio:
return (0, 0, 0)
winner = vals[idx]
return (int((winner >> 16) & 0xFF), int((winner >> 8) & 0xFF), int(winner & 0xFF))
# ---------------------------------------------------------------------------
# Checks
# ---------------------------------------------------------------------------
def check_inputs(pdf_path: Path, json_path: Path) -> tuple[fitz.Document, dict]:
assert pdf_path.exists(), f"PDF missing: {pdf_path}"
assert json_path.exists(), f"JSON missing: {json_path}"
doc = fitz.open(str(pdf_path))
parsed = json.loads(json_path.read_text(encoding="utf-8"))
assert "pages" in parsed, "parsed JSON missing 'pages'"
print(f" pdf pages = {doc.page_count}, parsed pages = {len(parsed['pages'])}")
assert len(parsed["pages"]) <= doc.page_count, (
f"page count mismatch — JSON has {len(parsed['pages'])} pages "
f"but PDF has {doc.page_count}"
)
return doc, parsed
def check_schema(parsed: dict) -> None:
cat_counter: Counter[str] = Counter()
label_counter: Counter[str] = Counter()
fontsize_zero = 0
cells_total = 0
cells_with_translated = 0
for page in parsed["pages"]:
for el in page["elements"]:
cat_counter[el["category"]] += 1
label_counter[el["label"]] += 1
assert "bbox_pdf" in el and len(el["bbox_pdf"]) == 4
if el["category"] != "BYPASS" and el.get("font_size", 0) == 0:
fontsize_zero += 1
for c in el.get("cells", []):
cells_total += 1
assert (
"source_text" in c and "bbox_pdf" in c
), "cell missing source_text or bbox_pdf — old fixture schema?"
if c.get("translated_text"):
cells_with_translated += 1
print(f" categories: {dict(cat_counter)}")
print(f" labels: {dict(label_counter)}")
print(f" zero font_size (non-BYPASS): {fontsize_zero}")
print(f" cells: {cells_total} total, {cells_with_translated} translated")
assert cells_total > 0, "no cells found — TABLE elements missing cells"
if cells_total and cells_with_translated == 0:
print(
" NOTE: no cells have translated_text (all-formula table or phase 2 not yet run)"
)
def check_overflow_risk(parsed: dict) -> None:
"""Plan-feasibility check: how often is JSON font_size > bbox height?
Validates the decision to treat font_size as an UPPER BOUND, not truth.
"""
risk_count = 0
total = 0
for page in parsed["pages"]:
for el in page["elements"]:
if el["category"] not in ("FLOWING_TEXT", "IN_PLACE"):
continue
total += 1
x0, y0, x1, y1 = el["bbox_pdf"]
h = y1 - y0
fs = el.get("font_size", 0) or 0
if fs > h:
risk_count += 1
pct = 100 * risk_count / max(1, total)
print(
f" font_size > bbox_height in {risk_count}/{total} ({pct:.0f}%) text elements"
)
if pct > 10:
print(
" WARN: > 10% overflow if font_size used as truth — shrink-to-fit REQUIRED"
)
def check_native_text(doc: fitz.Document) -> None:
"""Plan-feasibility check: scanned vs native PDF.
If the PDF has a native text layer, simple draw_rect erasure is insufficient —
we must use add_redact_annot + apply_redactions to remove the text layer.
"""
pages_with_text = 0
sample = ""
for i in range(min(3, doc.page_count)):
t = doc[i].get_text("text").strip()
if t:
pages_with_text += 1
if not sample:
sample = t[:80]
print(f" pages with native text (first 3): {pages_with_text}/3")
if sample:
print(f" sample: {sample!r}")
if pages_with_text:
print(" WARN: native text detected — redaction (not just rect erase) required")
def check_pixmap_and_color(doc: fitz.Document, parsed: dict) -> None:
page = doc[0]
pm = page.get_pixmap()
arr = _pixmap_to_array(pm)
print(f" pixmap {pm.width}×{pm.height} n={pm.n} → array {arr.shape} {arr.dtype}")
pw, ph = parsed["pages"][0]["page_width"], parsed["pages"][0]["page_height"]
# Pick the first non-BYPASS element with a non-trivial bbox.
target = None
for el in parsed["pages"][0]["elements"]:
if el["category"] != "BYPASS" and (el["bbox_pdf"][2] - el["bbox_pdf"][0]) > 50:
target = el
break
assert target is not None, "no testable element on page 0"
bbox_px = _bbox_to_pixels(target["bbox_pdf"], pw, ph, pm)
bg = detect_bg_color(arr, bbox_px)
txt = detect_text_color(arr, bbox_px, bg)
print(f" element label={target['label']} bbox_px={bbox_px}")
print(f" bg={bg} text={txt}")
assert all(0 <= c <= 255 for c in bg + txt), "color channel out of range"
def check_synthetic_render(font_path: Path | None) -> None:
"""Confirm fitz can register a font and insert non-trivial text into a page."""
fp = str(font_path) if (font_path and font_path.exists()) else None
doc = fitz.open()
page = doc.new_page(width=400, height=200)
if fp:
page.insert_font(fontname="Body", fontfile=fp)
fontname = "Body"
else:
fontname = "helv"
page.draw_rect(fitz.Rect(20, 20, 380, 60), color=(1, 1, 1), fill=(1, 1, 1))
rem = page.insert_textbox(
fitz.Rect(20, 20, 380, 60),
"Phase-3 smoke: ăn cơm chưa? (UTF-8 OK)" if fp else "Phase-3 smoke (no font)",
fontname=fontname,
fontsize=12,
color=(0, 0, 0),
align=fitz.TEXT_ALIGN_LEFT,
)
print(f" insert_textbox returned remaining={rem:.1f}")
assert rem >= 0, "even synthetic insert overflowed — fitz/font setup wrong"
extracted = page.get_text("text")
if fp:
normalized = extracted.replace("\xad", "-").replace("\xa0", " ")
assert "Phase-3" in normalized, f"text not extractable: {extracted!r}"
print(f" synthetic page text-extracts: {extracted.strip()[:60]!r}")
def check_render_module() -> bool:
try:
importlib.import_module("pdf2zh.render")
return True
except ModuleNotFoundError:
return False
def run_full_render(args) -> None:
from pdf2zh.render import RenderConfig, render_document # noqa: WPS433
cfg = RenderConfig(font_path=args.font)
cfg.keep_typst_source = True
parsed = json.loads(Path(args.parsed).read_text(encoding="utf-8"))
render_document(args.input, parsed, args.output, cfg)
out = Path(args.output)
assert out.exists() and out.stat().st_size > 0, "render produced empty file"
rdoc = fitz.open(str(out))
text0 = rdoc[0].get_text("text")
print(f" output {out.name}: {out.stat().st_size:,} bytes, {rdoc.page_count} pages")
print(f" page 0 extracted text excerpt: {text0.strip()[:120]!r}")
# Check that at least one translated element is findable in the output.
# Skip elements whose translated_text equals source_text (proper names, etc.)
first_translation = next(
(
el["translated_text"]
for el in parsed["pages"][0]["elements"]
if el["translated_text"]
and el["translated_text"] != el.get("source_text", "")
),
None,
)
if first_translation:
snippet = re.sub(r"<[^>]+>", "", first_translation)[:30].strip()
if snippet and snippet in text0:
print(f" translated snippet found in page 0: {snippet!r}")
else:
print(
f" NOTE: snippet not found in page 0 (may be covered by overlay): {snippet!r}"
)
# ---------------------------------------------------------------------------
# CLI
# ---------------------------------------------------------------------------
def main() -> int:
ap = argparse.ArgumentParser(
description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter
)
ap.add_argument("--input", required=True, help="Source PDF")
ap.add_argument("--parsed", required=True, help="Translated JSON (phase 2 output)")
ap.add_argument("--font", default=None, help="TTF font path (Unicode)")
ap.add_argument(
"--render", action="store_true", help="Run full render via pdf2zh.render"
)
ap.add_argument("--output", default=None, help="Output PDF path (when --render)")
args = ap.parse_args()
pdf_path = Path(args.input)
json_path = Path(args.parsed)
font_path = Path(args.font) if args.font else None
print("\n[1] check_inputs")
doc, parsed = check_inputs(pdf_path, json_path)
print("\n[2] check_schema")
check_schema(parsed)
print("\n[3] check_overflow_risk")
check_overflow_risk(parsed)
print("\n[4] check_native_text")
check_native_text(doc)
print("\n[5] check_pixmap_and_color")
check_pixmap_and_color(doc, parsed)
print("\n[6] check_synthetic_render")
check_synthetic_render(font_path)
print("\n[7] render module status: ", end="")
have_render = check_render_module()
print("AVAILABLE" if have_render else "not yet implemented (pdf2zh.render)")
if args.render:
if not have_render:
print(
"\nERROR: --render requested but pdf2zh.render module does not exist yet."
)
return 2
if not args.font or not args.output:
print("\nERROR: --render requires --font and --output.")
return 2
print("\n[8] run_full_render")
run_full_render(args)
print("\nALL CHECKS PASSED")
return 0
if __name__ == "__main__":
sys.exit(main())
|