UFR-Fing / scripts /eda_nist302_inputs.py
anbinh39's picture
Upload UFR-Fing converted data and manifests
73dcd2c verified
Raw
History Blame Contribute Delete
9.02 kB
from __future__ import annotations
import argparse
import csv
import json
import re
from collections import Counter, defaultdict
from pathlib import Path
from statistics import mean
from typing import Any
FS = "\x1c"
GS = "\x1d"
RS = "\x1e"
US = "\x1f"
def read_text_prefix(path: Path, max_bytes: int = 2_000_000) -> str:
data = path.read_bytes()[:max_bytes]
return data.decode("latin-1", errors="replace")
def split_ebts_records(path: Path) -> dict[str, list[dict[str, str]]]:
text = read_text_prefix(path)
png_idx = text.find("\x89PNG\r\n\x1a\n")
if png_idx >= 0:
text = text[:png_idx]
records: dict[str, list[dict[str, str]]] = defaultdict(list)
for record in text.split(FS):
record = record.strip("\x00\r\n")
if not record:
continue
fields: dict[str, str] = {}
for field in record.split(GS):
if ":" not in field:
continue
key, value = field.split(":", 1)
if "." not in key:
continue
record_type = key.split(".", 1)[0]
fields[key] = value
if fields:
record_type = next(iter(fields)).split(".", 1)[0]
records[record_type].append(fields)
return dict(records)
def count_subfields(value: str | None) -> int:
if not value:
return 0
return len([part for part in value.split(RS) if part])
def segmentation_grid_shape(value: str | None) -> tuple[int | None, int | None, set[str]]:
if not value:
return None, None, set()
rows = [row for row in value.split(RS) if row]
widths = {len(row) for row in rows}
chars = set("".join(rows))
width = widths.pop() if len(widths) == 1 else None
return len(rows), width, chars
def parse_int_pair(value: str | None) -> tuple[int | None, int | None]:
if not value:
return None, None
parts = value.split(US)
if len(parts) < 2:
return None, None
try:
return int(parts[0]), int(parts[1])
except ValueError:
return None, None
def png_ihdr_from_text(text: str) -> tuple[int | None, int | None]:
marker = "\x89PNG\r\n\x1a\n"
idx = text.find(marker)
if idx < 0:
return None, None
raw = text.encode("latin-1", errors="replace")
pos = idx + len(marker)
if len(raw) < pos + 16:
return None, None
# IHDR chunk: length(4), type(4), width(4), height(4)
if raw[pos + 4 : pos + 8] != b"IHDR":
return None, None
width = int.from_bytes(raw[pos + 8 : pos + 12], "big")
height = int.from_bytes(raw[pos + 12 : pos + 16], "big")
return width, height
def comp_references(path: Path) -> tuple[str | None, str | None]:
text = read_text_prefix(path, 400_000)
lffs = None
irr = None
# Field 2.1406 contains references with labels LFFS/IRR in the current data.
m = re.search(r"LFFS" + US + r"[^" + RS + FS + r"]*" + US + r"([^" + US + RS + FS + r"]+\.lffs)", text)
if m:
lffs = Path(m.group(1)).name
m = re.search(r"IRR" + US + r"[^" + RS + FS + r"]*" + US + r"([^" + US + RS + FS + r"]+\.irr)", text)
if m:
irr = Path(m.group(1)).name
return lffs, irr
def summarize_ebts_file(path: Path) -> dict[str, Any]:
records = split_ebts_records(path)
out: dict[str, Any] = {
"record_types": {record_type: len(items) for record_type, items in sorted(records.items())},
"type_1_file_content": records.get("1", [{}])[0].get("1.003"),
"type_1_type_of_transaction": records.get("1", [{}])[0].get("1.004"),
}
type9 = records.get("9", [])
out["type9_count"] = len(type9)
out["type13_count"] = len(records.get("13", []))
out["type9_minutiae_counts"] = [count_subfields(r.get("9.331")) for r in type9]
out["type9_core_counts"] = [count_subfields(r.get("9.320")) for r in type9]
out["type9_delta_counts"] = [count_subfields(r.get("9.321")) for r in type9]
out["type9_seg_shapes"] = []
out["type9_image_sizes"] = []
out["type9_imp"] = []
for r in type9:
seg_h, seg_w, chars = segmentation_grid_shape(r.get("9.308"))
out["type9_seg_shapes"].append([seg_w, seg_h, "".join(sorted(chars))])
out["type9_image_sizes"].append(list(parse_int_pair(r.get("9.300"))))
out["type9_imp"].append(r.get("9.004"))
type13 = records.get("13", [])
out["type13_image_sizes"] = []
out["type13_hll_vll"] = []
for r in type13:
out["type13_hll_vll"].append([r.get("13.006"), r.get("13.007"), r.get("13.009"), r.get("13.010")])
text = read_text_prefix(path, 2_000_000)
out["embedded_png_ihdr"] = list(png_ihdr_from_text(text))
if path.suffix == ".comp":
out["comp_references"] = list(comp_references(path))
return out
def counter_to_dict(counter: Counter[Any]) -> dict[str, int]:
return {str(k): int(v) for k, v in counter.most_common()}
def summarize_manifest(path: Path) -> dict[str, Any]:
with path.open(newline="", encoding="utf-8") as handle:
rows = list(csv.DictReader(handle))
out: dict[str, Any] = {"path": str(path), "rows": len(rows), "columns": list(rows[0].keys()) if rows else []}
for col in ("latent", "mate", "mate_irr", "comp_path", "lffs_path", "identity_label", "subject", "fgp", "status"):
if rows and col in rows[0]:
vals = [r.get(col, "") for r in rows]
out[f"{col}_nonempty"] = sum(1 for v in vals if v)
out[f"{col}_unique"] = len(set(v for v in vals if v))
if col in ("status", "fgp"):
out[f"{col}_counts"] = counter_to_dict(Counter(vals))
return out
def summarize_image_csv(path: Path, limit: int = 300) -> dict[str, Any]:
try:
from PIL import Image
except Exception:
return {"path": str(path), "status": "PIL_unavailable"}
with path.open(newline="", encoding="utf-8") as handle:
rows = list(csv.DictReader(handle))
stats: dict[str, list[float]] = defaultdict(list)
for row in rows[:limit]:
for col in ("latent", "mate"):
image_path = row.get(col)
if not image_path:
continue
p = Path(image_path)
if not p.exists():
continue
try:
im = Image.open(p)
except Exception:
continue
stats[f"{col}_width"].append(float(im.size[0]))
stats[f"{col}_height"].append(float(im.size[1]))
dpi = im.info.get("dpi")
if dpi:
stats[f"{col}_dpi_x"].append(float(dpi[0]))
stats[f"{col}_dpi_y"].append(float(dpi[1]))
return {
"path": str(path),
"sampled_rows": min(limit, len(rows)),
"stats": {
key: {
"min": min(values),
"mean": mean(values),
"max": max(values),
"unique": sorted(set(values))[:20],
}
for key, values in stats.items()
},
}
def main() -> int:
parser = argparse.ArgumentParser()
parser.add_argument("--dataset-root", default="/home/aiserver/works/fingerprint/dataset")
parser.add_argument("--manifest-root", default="manifests/nist302")
parser.add_argument("--out", default="outputs/eda_nist302_inputs.json")
parser.add_argument("--sample-files", type=int, default=20)
args = parser.parse_args()
dataset_root = Path(args.dataset_root)
manifest_root = Path(args.manifest_root)
out_path = Path(args.out)
report: dict[str, Any] = {}
nist_dirs = sorted(p for p in dataset_root.iterdir() if p.is_dir() and p.name.startswith("nist302"))
report["dataset_roots"] = [str(p) for p in nist_dirs]
report["extension_counts_by_root"] = {}
for root in nist_dirs:
counter: Counter[str] = Counter()
for path in root.rglob("*"):
if path.is_file():
counter[path.suffix.lower() or "<none>"] += 1
report["extension_counts_by_root"][root.name] = counter_to_dict(counter)
manifest_files = sorted(manifest_root.rglob("*.csv"))
report["manifests"] = [summarize_manifest(p) for p in manifest_files]
ready_manifests = sorted(manifest_root.glob("*_ready/paired_302i_*.csv"))
report["ready_manifest_image_stats_sample"] = [summarize_image_csv(p) for p in ready_manifests]
ebts_samples: dict[str, list[dict[str, Any]]] = {}
for suffix in (".comp", ".lffs", ".irr"):
files = sorted(dataset_root.rglob(f"*{suffix}"))[: args.sample_files]
ebts_samples[suffix] = [{"path": str(p), **summarize_ebts_file(p)} for p in files]
report["ebts_samples"] = ebts_samples
out_path.parent.mkdir(parents=True, exist_ok=True)
out_path.write_text(json.dumps(report, indent=2, sort_keys=True) + "\n", encoding="utf-8")
print(json.dumps({"out": str(out_path), "dataset_roots": len(nist_dirs), "manifests": len(manifest_files)}, indent=2))
return 0
if __name__ == "__main__":
raise SystemExit(main())