File size: 6,003 Bytes
d82db76 | 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 | from __future__ import annotations
import importlib.util
import json
import sys
from argparse import Namespace
from pathlib import Path
from ocrtech.manifest import ManifestEntry, sha256_file, write_manifest
from ocrtech.schemas import BBox, Block, Document, Page, TextLine
def _load_tool():
path = Path(__file__).resolve().parents[1] / "tools" / "audit_page_ocr_documents.py"
spec = importlib.util.spec_from_file_location("audit_page_ocr_documents", path)
assert spec is not None
assert spec.loader is not None
module = importlib.util.module_from_spec(spec)
sys.modules[spec.name] = module
spec.loader.exec_module(module)
return module
def _write_fixture(tmp_path: Path, *, copy_eval_image: bool = False) -> tuple[Path, Path]:
source_image = tmp_path / "prepared.png"
source_image.write_bytes(b"same prepared image bytes")
eval_image = tmp_path / "prepared-copy.png" if copy_eval_image else source_image
if copy_eval_image:
eval_image.write_bytes(source_image.read_bytes())
crop = tmp_path / "crop.png"
crop.write_bytes(b"line crop bytes")
manifest = tmp_path / "manifest.jsonl"
write_manifest(
[
ManifestEntry(
sample_id="page-1",
dataset="unit-pages",
split="eval",
image_path=str(eval_image),
text="ᤁᤂᤃ",
sha256=sha256_file(eval_image),
)
],
manifest,
)
document = Document(
source_path=str(source_image),
pages=[
Page(
page_index=0,
width=100,
height=120,
text_lines=[
TextLine(
text="ᤁᤂᤃ",
bbox=BBox(10, 12, 50, 14),
confidence=0.91,
page_index=0,
line_id="line-1",
metadata={
"detector_source_pass": "yolo-model",
"structural_role": "text",
"crop_path": str(crop),
"crop_sha256": sha256_file(crop),
},
)
],
blocks=[
Block(
block_id="block-1",
block_type="text",
page_index=0,
bbox=BBox(10, 12, 50, 14),
order=0,
text="ᤁᤂᤃ",
line_ids=["line-1"],
)
],
)
],
metadata={"ocr_output_kind": "live_detector", "line_count": 1},
)
docs = tmp_path / "documents"
sample_dir = docs / "page-1"
sample_dir.mkdir(parents=True)
(sample_dir / "document.json").write_text(
json.dumps(document.to_dict(), ensure_ascii=False, indent=2) + "\n",
encoding="utf-8",
)
return manifest, docs
def _args(manifest: Path, docs: Path, **overrides) -> Namespace:
values = {
"eval_manifest": manifest,
"documents_dir": docs,
"out": None,
"expected_ocr_output_kind": "live_detector",
"expected_detector_source_pass": "yolo-model",
"require_source_image": True,
"require_document_source_match": True,
"require_line_crops": True,
"require_line_confidence": True,
"allow_empty_text": False,
"bbox_tolerance_px": 2.0,
"min_line_count": 1,
"max_line_count": None,
}
values.update(overrides)
return Namespace(**values)
def test_audit_page_ocr_documents_writes_per_page_output_audits(tmp_path: Path) -> None:
tool = _load_tool()
manifest, docs = _write_fixture(tmp_path, copy_eval_image=True)
result = tool.audit_page_ocr_documents(_args(manifest, docs))
assert result["passed"] is True
assert result["passed_count"] == 1
sample_audit = docs / "page-1" / "limbu-output-audit.json"
assert sample_audit.is_file()
payload = json.loads(sample_audit.read_text(encoding="utf-8"))
assert payload["passed"] is True
assert payload["source_binding"]["match_kind"] == "same_sha256"
assert payload["line_count"] == 1
assert payload["crop_hash_checked_count"] == 1
assert (docs / "page-ocr-documents-audit.json").is_file()
def test_audit_page_ocr_documents_rejects_source_image_mismatch(tmp_path: Path) -> None:
tool = _load_tool()
manifest, docs = _write_fixture(tmp_path)
document_path = docs / "page-1" / "document.json"
payload = json.loads(document_path.read_text(encoding="utf-8"))
wrong = tmp_path / "wrong.png"
wrong.write_bytes(b"different image")
payload["source_path"] = str(wrong)
document_path.write_text(json.dumps(payload, ensure_ascii=False, indent=2) + "\n", encoding="utf-8")
result = tool.audit_page_ocr_documents(_args(manifest, docs))
assert result["passed"] is False
assert result["failed_count"] == 1
sample = result["samples"][0]
assert any("does not match eval input by path or SHA-256" in issue for issue in sample["issues"])
assert json.loads((docs / "page-1" / "limbu-output-audit.json").read_text(encoding="utf-8"))["passed"] is False
def test_audit_page_ocr_documents_rejects_tampered_crop_hash(tmp_path: Path) -> None:
tool = _load_tool()
manifest, docs = _write_fixture(tmp_path)
document_path = docs / "page-1" / "document.json"
payload = json.loads(document_path.read_text(encoding="utf-8"))
payload["pages"][0]["text_lines"][0]["metadata"]["crop_sha256"] = "0" * 64
document_path.write_text(json.dumps(payload, ensure_ascii=False, indent=2) + "\n", encoding="utf-8")
result = tool.audit_page_ocr_documents(_args(manifest, docs))
assert result["passed"] is False
assert any("crop_sha256 mismatch" in issue for issue in result["samples"][0]["issues"])
|