[verified] generalize B1 response regions and marks gap fill
api-ci-deploy / test-build-deploy (push) Has been cancelled

This commit is contained in:
2026-06-08 04:49:21 +01:00
parent 69d9c46abe
commit 52d1ece212
4 changed files with 179 additions and 15 deletions
+16 -7
View File
@@ -67,49 +67,49 @@ B1_GEOMETRY = [
pdf="samples/b1/aqa-biology-7402-1-2023jun.pdf",
docling="results/b1_rapid/b1-aqa-biology-7402-1-2023jun/merged.json",
rapid="results/b1_rapid/b1-aqa-biology-7402-1-2023jun/p*.json",
gt_key="b1-aqa-biology-7402-1-2023jun"),
gt_key="b1-aqa-biology-7402-1-2023jun", expected_max=91),
dict(slug="b1-aqa-chemistry-7405-1-2022jun", title="AQA A-level Chemistry 7405/1 2022 Jun (image-only OCR baseline)",
board="aqa", level="A-level", path="B1 image-only OCR (RapidOCR margin-pass)",
storage_loc="cc.examboards/aqa/chemistry/7405/1/2022-jun/qp.pdf",
pdf="samples/b1/aqa-chemistry-7405-1-2022jun.pdf",
docling="results/b1_rapid/b1-aqa-chemistry-7405-1-2022jun/merged.json",
rapid="results/b1_rapid/b1-aqa-chemistry-7405-1-2022jun/p*.json",
gt_key="b1-aqa-chemistry-7405-1-2022jun"),
gt_key="b1-aqa-chemistry-7405-1-2022jun", expected_max=105),
dict(slug="b1-aqa-physics-7408-1-2022jun", title="AQA A-level Physics 7408/1 2022 Jun (image-only OCR baseline)",
board="aqa", level="A-level", path="B1 image-only OCR (RapidOCR margin-pass)",
storage_loc="cc.examboards/aqa/physics/7408/1/2022-jun/qp.pdf",
pdf="samples/b1/aqa-physics-7408-1-2022jun.pdf",
docling="results/b1_rapid/b1-aqa-physics-7408-1-2022jun/merged.json",
rapid="results/b1_rapid/b1-aqa-physics-7408-1-2022jun/p*.json",
gt_key="b1-aqa-physics-7408-1-2022jun"),
gt_key="b1-aqa-physics-7408-1-2022jun", expected_max=85),
dict(slug="b1-aqa-biology-8461-1h-2022jun", title="AQA GCSE Biology 8461/1H 2022 Jun (image-only OCR baseline)",
board="aqa", level="GCSE", path="B1 image-only OCR (RapidOCR margin-pass)",
storage_loc="cc.examboards/aqa/biology/8461/1h/2022-jun/qp.pdf",
pdf="samples/b1/aqa-biology-8461-1h-2022jun.pdf",
docling="results/b1_rapid/b1-aqa-biology-8461-1h-2022jun/merged.json",
rapid="results/b1_rapid/b1-aqa-biology-8461-1h-2022jun/p*.json",
gt_key="b1-aqa-biology-8461-1h-2022jun"),
gt_key="b1-aqa-biology-8461-1h-2022jun", expected_max=100),
dict(slug="b1-aqa-chemistry-8462-1h-2022jun", title="AQA GCSE Chemistry 8462/1H 2022 Jun (image-only OCR baseline)",
board="aqa", level="GCSE", path="B1 image-only OCR (RapidOCR margin-pass)",
storage_loc="cc.examboards/aqa/chemistry/8462/1h/2022-jun/qp.pdf",
pdf="samples/b1/aqa-chemistry-8462-1h-2022jun.pdf",
docling="results/b1_rapid/b1-aqa-chemistry-8462-1h-2022jun/merged.json",
rapid="results/b1_rapid/b1-aqa-chemistry-8462-1h-2022jun/p*.json",
gt_key="b1-aqa-chemistry-8462-1h-2022jun"),
gt_key="b1-aqa-chemistry-8462-1h-2022jun", expected_max=100),
dict(slug="b1-aqa-combined-8464-b1h-2022jun", title="AQA GCSE Combined Science Trilogy 8464/B/1H 2022 Jun (image-only OCR baseline)",
board="aqa", level="GCSE", path="B1 image-only OCR (RapidOCR margin-pass)",
storage_loc="cc.examboards/aqa/combined-science-trilogy/8464/b-1h/2022-jun/qp.pdf",
pdf="samples/b1/aqa-combined-8464-b1h-2022jun.pdf",
docling="results/b1_rapid/b1-aqa-combined-8464-b1h-2022jun/merged.json",
rapid="results/b1_rapid/b1-aqa-combined-8464-b1h-2022jun/p*.json",
gt_key="b1-aqa-combined-8464-b1h-2022jun"),
gt_key="b1-aqa-combined-8464-b1h-2022jun", expected_max=70),
dict(slug="b1-aqa-combined-8464-c1h-2022jun", title="AQA GCSE Combined Science Trilogy 8464/C/1H 2022 Jun (image-only OCR baseline; 8465 not present in dev catalogue)",
board="aqa", level="GCSE", path="B1 image-only OCR (RapidOCR margin-pass)",
storage_loc="cc.examboards/aqa/combined-science-trilogy/8464/c-1h/2022-jun/qp.pdf",
pdf="samples/b1/aqa-combined-8464-c1h-2022jun.pdf",
docling="results/b1_rapid/b1-aqa-combined-8464-c1h-2022jun/merged.json",
rapid="results/b1_rapid/b1-aqa-combined-8464-c1h-2022jun/p*.json",
gt_key="b1-aqa-combined-8464-c1h-2022jun"),
gt_key="b1-aqa-combined-8464-c1h-2022jun", expected_max=70),
]
GT_LABELS_PATH = "fixtures/b1_gt_labels.json"
@@ -209,6 +209,9 @@ def stats_from(struct, val, gt_labels=None):
"coverage_pct": cov.get("coverage_pct"), "coverage_recovered": cov.get("recovered"),
"coverage_total": cov.get("total"), "coverage_source": cov.get("source"),
"coverage_missed": cov.get("missed", []), "answer_regions": answer_region_count(struct),
"opencv_answer_regions": st.get("opencv_answer_regions"),
"opencv_answer_region_candidates": st.get("opencv_answer_region_candidates"),
"residual_marks_gapfilled": st.get("residual_marks_gapfilled"),
"validate_verdict": (val.get("summary") or {}).get("worst_severity"),
"validate_flags": val.get("flags", []),
"questions_expected": (val.get("summary") or {}).get("questions_expected"),
@@ -226,6 +229,10 @@ def do_geometry(p, overlays, gt_labels=None, prepare_ocr=False):
raise RuntimeError(f"unable to prepare B1 OCR cache for {p['slug']}")
extract_args = p.get("extract") or ["--docling", p["docling"], "--rapid", p["rapid"], "--board", p.get("board", "aqa")]
ex = ["extract.py"] + extract_args + ["--out", S]
if p.get("pdf"):
ex += ["--response-regions", p["pdf"]]
if p.get("expected_max"):
ex += ["--expected-max", str(p["expected_max"])]
if p.get("gt"):
ex += ["--gt", p["gt"]]
run(ex)
@@ -272,6 +279,8 @@ def per_paper_report(p, s, d, kind):
if s['coverage_pct'] is not None else "- **coverage vs GT:** n/a",
f"- **G6 verdict:** {s['validate_verdict']}",
f"- **answer-region count:** {s.get('answer_regions')}",
f"- **opencv response regions:** {s.get('opencv_answer_regions')} attached / "
f"{s.get('opencv_answer_region_candidates')} candidates",
]
if s["validate_flags"]:
lines += ["", "**Flags (human-review hints):**"] + [f"- {f}" for f in s["validate_flags"]]