spectre_pdf 1.0.0

Native Rust PDF extraction engine: text, markdown for RAG, AcroForm widgets, image decoding, and encrypted PDFs. Lazy parser, persistent Document handle, no C dependencies.
Documentation 
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

"""
Throwaway: heading-F1 sweep across a diverse PDF set with non-empty TOCs.

Imports the scoring primitives from the existing bench script so we keep
parity with the i1040gi baseline measurement.
"""
from __future__ import annotations
import json
import sys
import time
from pathlib import Path

sys.path.insert(0, "/mnt/c/Users/RyanJ/spectre-rs/scripts")
from bench_markdown_heading_f1 import (  # type: ignore
    normalize_heading,
    headings_from_markdown,
    prec_recall_f1,
)


def headings_from_toc(toc):
    """Override: bench helper has a precedence bug — `getattr(entry, "title", None) or entry[1]`
    triggers `entry[1]` when title is an empty string, which crashes on TocEntry objects."""
    out = set()
    for entry in toc:
        if hasattr(entry, "title"):
            title = entry.title
        elif isinstance(entry, dict):
            title = entry.get("title", "")
        else:
            title = entry[1] if len(entry) > 1 else ""
        h = normalize_heading(title or "")
        if h and len(h) > 1:
            out.add(h)
    return out

import spectre_rs
import pymupdf4llm

CACHE_PATH = Path("/tmp/spectre_extra_corpus/pmu4llm_cache.json")

CORPUS = [
    # (label, path)
    ("i1040gi",       "/mnt/c/Users/RyanJ/spectre-rs/bench-corpus/markdown/i1040gi.pdf"),
    ("p17",           "/tmp/spectre_extra_corpus/p17.pdf"),
    ("p334",          "/tmp/spectre_extra_corpus/p334.pdf"),
    ("p550",          "/tmp/spectre_extra_corpus/p550.pdf"),
    ("p525",          "/tmp/spectre_extra_corpus/p525.pdf"),
    ("p590a",         "/tmp/spectre_extra_corpus/p590a.pdf"),
    ("p15",           "/tmp/spectre_extra_corpus/p15.pdf"),
    ("p463",          "/tmp/spectre_extra_corpus/p463.pdf"),
    ("freeculture",   "/mnt/c/Users/RyanJ/spectre-rs/bench-corpus/pdfjs/pdf.js-master/test/pdfs/freeculture.pdf"),
    ("arxiv_lora",    "/tmp/spectre_extra_corpus/2106.09685.pdf"),
    ("arxiv_mineru",  "/tmp/spectre_extra_corpus/2409.18839.pdf"),
    ("arxiv_tot",     "/tmp/spectre_extra_corpus/2305.10601.pdf"),
    ("arxiv_mistral", "/tmp/spectre_extra_corpus/2310.06825.pdf"),
    ("issue10900",    "/mnt/c/Users/RyanJ/spectre-rs/bench-corpus/pdfjs/pdf.js-master/test/pdfs/issue10900.pdf"),
    ("xfa_imm1344e",  "/mnt/c/Users/RyanJ/spectre-rs/bench-corpus/pdfjs/pdf.js-master/test/pdfs/xfa_filled_imm1344e.pdf"),
    ("file_pdfjs",    "/mnt/c/Users/RyanJ/spectre-rs/bench-corpus/pdfjs/pdf.js-master/test/pdfs/file_pdfjs_test.pdf"),
    ("labelled_pages","/mnt/c/Users/RyanJ/spectre-rs/bench-corpus/pdfjs/pdf.js-master/test/pdfs/labelled_pages.pdf"),
    ("nested_outline","/mnt/c/Users/RyanJ/spectre-rs/bench-corpus/pdfjs/pdf.js-master/test/pdfs/nested_outline.pdf"),
    ("icdar_us008",   "/mnt/c/Users/RyanJ/spectre-rs/bench-corpus/ICDAR_2013_table_evaluate-master/icdar2013-competition-dataset-with-gt/pdf/us-008.pdf"),
    ("icdar_us036",   "/mnt/c/Users/RyanJ/spectre-rs/bench-corpus/ICDAR_2013_table_evaluate-master/icdar2013-competition-dataset-with-gt/pdf/us-036.pdf"),
    # IRS publications (heading-detection corpus extension, 30+ docs).
    ("p1",     "/tmp/spectre_extra_corpus/p1.pdf"),
    ("p3",     "/tmp/spectre_extra_corpus/p3.pdf"),
    ("p15a",   "/tmp/spectre_extra_corpus/p15a.pdf"),
    ("p15b",   "/tmp/spectre_extra_corpus/p15b.pdf"),
    ("p225",   "/tmp/spectre_extra_corpus/p225.pdf"),
    ("p501",   "/tmp/spectre_extra_corpus/p501.pdf"),
    ("p502",   "/tmp/spectre_extra_corpus/p502.pdf"),
    ("p503",   "/tmp/spectre_extra_corpus/p503.pdf"),
    ("p504",   "/tmp/spectre_extra_corpus/p504.pdf"),
    ("p505",   "/tmp/spectre_extra_corpus/p505.pdf"),
    ("p514",   "/tmp/spectre_extra_corpus/p514.pdf"),
    ("p526",   "/tmp/spectre_extra_corpus/p526.pdf"),
    ("p527",   "/tmp/spectre_extra_corpus/p527.pdf"),
    ("p535",   "/tmp/spectre_extra_corpus/p535.pdf"),
    ("p536",   "/tmp/spectre_extra_corpus/p536.pdf"),
    ("p544",   "/tmp/spectre_extra_corpus/p544.pdf"),
    ("p551",   "/tmp/spectre_extra_corpus/p551.pdf"),
    ("p554",   "/tmp/spectre_extra_corpus/p554.pdf"),
    ("p556",   "/tmp/spectre_extra_corpus/p556.pdf"),
    ("p559",   "/tmp/spectre_extra_corpus/p559.pdf"),
    ("p560",   "/tmp/spectre_extra_corpus/p560.pdf"),
    ("p575",   "/tmp/spectre_extra_corpus/p575.pdf"),
    ("p590b",  "/tmp/spectre_extra_corpus/p590b.pdf"),
    ("p721",   "/tmp/spectre_extra_corpus/p721.pdf"),
    ("p908",   "/tmp/spectre_extra_corpus/p908.pdf"),
    ("p915",   "/tmp/spectre_extra_corpus/p915.pdf"),
    ("p925",   "/tmp/spectre_extra_corpus/p925.pdf"),
    ("p926",   "/tmp/spectre_extra_corpus/p926.pdf"),
    ("p936",   "/tmp/spectre_extra_corpus/p936.pdf"),
    ("p946",   "/tmp/spectre_extra_corpus/p946.pdf"),
    ("p969",   "/tmp/spectre_extra_corpus/p969.pdf"),
    ("p970",   "/tmp/spectre_extra_corpus/p970.pdf"),
    ("p974",   "/tmp/spectre_extra_corpus/p974.pdf"),
]


def load_cache() -> dict:
    if CACHE_PATH.exists():
        return json.loads(CACHE_PATH.read_text())
    return {}


def save_cache(c: dict) -> None:
    CACHE_PATH.write_text(json.dumps(c, indent=2))


def main() -> None:
    cache = load_cache()
    rows = []
    t_total = time.time()
    for label, path in CORPUS:
        p = Path(path)
        if not p.exists():
            print(f"MISS {label} {path}", file=sys.stderr)
            continue
        b = p.read_bytes()
        try:
            d = spectre_rs.Document(b)
            toc = d.toc()
        except Exception as e:
            print(f"FAIL {label}: spectre load — {e}", file=sys.stderr)
            continue
        gt = headings_from_toc(toc)
        if not gt:
            print(f"SKIP {label} (no usable TOC)", file=sys.stderr)
            continue

        # spectre
        t0 = time.time()
        spec_md = d.markdown()
        spec_t = time.time() - t0
        spec_h = headings_from_markdown(spec_md)
        s_p, s_r, s_f1 = prec_recall_f1(spec_h, gt)

        # pymupdf4llm — cache by path + mtime
        key = f"{path}|{p.stat().st_mtime_ns}"
        if key in cache:
            entry = cache[key]
            pmu_t = entry.get("seconds", -1.0)
            pmu_h_count = entry["pmu_h_count"]
            p_p, p_r, p_f1 = entry["p"], entry["r"], entry["f1"]
            cached = True
        else:
            t0 = time.time()
            pmu_md = pymupdf4llm.to_markdown(path)
            pmu_t = time.time() - t0
            pmu_h = headings_from_markdown(pmu_md)
            p_p, p_r, p_f1 = prec_recall_f1(pmu_h, gt)
            pmu_h_count = len(pmu_h)
            cache[key] = {
                "seconds": pmu_t,
                "pmu_h_count": pmu_h_count,
                "p": p_p, "r": p_r, "f1": p_f1,
            }
            save_cache(cache)
            cached = False

        row = {
            "label": label,
            "toc": len(gt),
            "spec_h": len(spec_h),
            "pmu_h": pmu_h_count,
            "spec_p": s_p, "spec_r": s_r, "spec_f1": s_f1,
            "pmu_p": p_p, "pmu_r": p_r, "pmu_f1": p_f1,
            "spec_seconds": spec_t,
            "pmu_seconds": pmu_t,
            "pmu_cached": cached,
        }
        rows.append(row)
        print(f"OK   {label:18s} TOC={len(gt):4d} "
              f"spec F1={s_f1:.2f} ({spec_t:5.2f}s) "
              f"pmu F1={p_f1:.2f} ({'cache' if cached else f'{pmu_t:5.1f}s'})",
              file=sys.stderr)

    out = {"rows": rows, "wall_seconds": time.time() - t_total}
    print(json.dumps(out, indent=2))


if __name__ == "__main__":
    main()