scripts/eda_hq_image_groups.py

from __future__ import annotations

import argparse
import csv
import hashlib
import json
import math
import re
from collections import Counter, defaultdict
from pathlib import Path
from statistics import mean, median
from typing import Any

import numpy as np
from PIL import Image


def percentile(values: list[float], q: float) -> float | None:
    if not values:
        return None
    return float(np.percentile(np.asarray(values, dtype=np.float64), q))


def describe(values: list[float]) -> dict[str, Any]:
    if not values:
        return {}
    return {
        "count": len(values),
        "min": float(min(values)),
        "p05": percentile(values, 5),
        "mean": float(mean(values)),
        "median": float(median(values)),
        "p95": percentile(values, 95),
        "max": float(max(values)),
    }


def counter_dict(counter: Counter[Any], limit: int = 30) -> dict[str, int]:
    return {str(key): int(value) for key, value in counter.most_common(limit)}


def file_sha256(path: Path) -> str:
    digest = hashlib.sha256()
    with path.open("rb") as handle:
        for chunk in iter(lambda: handle.read(1024 * 1024), b""):
            digest.update(chunk)
    return digest.hexdigest()


def image_entropy(arr: np.ndarray) -> float:
    hist = np.bincount(arr.reshape(-1), minlength=256).astype(np.float64)
    probs = hist[hist > 0] / hist.sum()
    return float(-(probs * np.log2(probs)).sum())


def laplacian_variance(arr: np.ndarray) -> float:
    if min(arr.shape) < 3:
        return 0.0
    center = arr[1:-1, 1:-1].astype(np.float32)
    lap = (
        arr[:-2, 1:-1].astype(np.float32)
        + arr[2:, 1:-1].astype(np.float32)
        + arr[1:-1, :-2].astype(np.float32)
        + arr[1:-1, 2:].astype(np.float32)
        - 4.0 * center
    )
    return float(lap.var())


def downsample_for_stats(image: Image.Image, max_side: int = 768) -> Image.Image:
    width, height = image.size
    scale = min(1.0, max_side / max(width, height))
    if scale >= 1.0:
        return image
    size = (max(1, int(round(width * scale))), max(1, int(round(height * scale))))
    return image.resize(size, Image.Resampling.BILINEAR)


def image_record(path: Path, hash_files: bool) -> dict[str, Any]:
    with Image.open(path) as image:
        mode = image.mode
        fmt = image.format
        width, height = image.size
        dpi = image.info.get("dpi")
        gray = downsample_for_stats(image.convert("L"))
        arr = np.asarray(gray, dtype=np.uint8)

    foreground = arr < 245
    if foreground.any():
        ys, xs = np.where(foreground)
        bbox_area_ratio = float(((xs.max() - xs.min() + 1) * (ys.max() - ys.min() + 1)) / arr.size)
    else:
        bbox_area_ratio = 0.0

    p05 = float(np.percentile(arr, 5))
    p95 = float(np.percentile(arr, 95))
    out = {
        "path": str(path),
        "mode": mode,
        "format": fmt,
        "width": width,
        "height": height,
        "area": width * height,
        "aspect": width / height if height else None,
        "dpi_x": float(dpi[0]) if dpi else None,
        "dpi_y": float(dpi[1]) if dpi else None,
        "mean": float(arr.mean()),
        "std": float(arr.std()),
        "p01": float(np.percentile(arr, 1)),
        "p05": p05,
        "p50": float(np.percentile(arr, 50)),
        "p95": p95,
        "p99": float(np.percentile(arr, 99)),
        "contrast_p95_p05": p95 - p05,
        "black_frac_lte_5": float((arr <= 5).mean()),
        "dark_frac_lt_80": float((arr < 80).mean()),
        "ink_frac_lt_245": float(foreground.mean()),
        "white_frac_gte_250": float((arr >= 250).mean()),
        "entropy": image_entropy(arr),
        "laplacian_var": laplacian_variance(arr),
        "foreground_bbox_area_ratio": bbox_area_ratio,
        "sha256": file_sha256(path) if hash_files else None,
    }
    return out


def metadata_record(path: Path, hash_files: bool) -> dict[str, Any]:
    with Image.open(path) as image:
        dpi = image.info.get("dpi")
        return {
            "path": str(path),
            "mode": image.mode,
            "format": image.format,
            "width": image.size[0],
            "height": image.size[1],
            "area": image.size[0] * image.size[1],
            "aspect": image.size[0] / image.size[1] if image.size[1] else None,
            "dpi_x": float(dpi[0]) if dpi else None,
            "dpi_y": float(dpi[1]) if dpi else None,
            "sha256": file_sha256(path) if hash_files else None,
        }


def even_sample(paths: list[Path], limit: int | None) -> list[Path]:
    if limit is None or limit <= 0 or len(paths) <= limit:
        return paths
    if limit == 1:
        return [paths[0]]
    indexes = np.linspace(0, len(paths) - 1, limit, dtype=np.int64)
    return [paths[int(index)] for index in indexes]


def summarize_images(
    paths: list[Path],
    sample_records: int,
    hash_files: bool,
    quality_limit: int | None,
) -> dict[str, Any]:
    metadata: list[dict[str, Any]] = []
    errors: list[dict[str, str]] = []
    for path in paths:
        try:
            metadata.append(metadata_record(path, hash_files))
        except Exception as exc:
            errors.append({"path": str(path), "error": repr(exc)})

    quality_paths = even_sample([Path(item["path"]) for item in metadata], quality_limit)
    quality_records: list[dict[str, Any]] = []
    quality_errors: list[dict[str, str]] = []
    for path in quality_paths:
        try:
            quality_records.append(image_record(path, False))
        except Exception as exc:
            quality_errors.append({"path": str(path), "error": repr(exc)})

    metadata_numeric_keys = [
        "width",
        "height",
        "area",
        "aspect",
        "dpi_x",
        "dpi_y",
    ]
    quality_numeric_keys = [
        "mean",
        "std",
        "p01",
        "p05",
        "p50",
        "p95",
        "p99",
        "contrast_p95_p05",
        "black_frac_lte_5",
        "dark_frac_lt_80",
        "ink_frac_lt_245",
        "white_frac_gte_250",
        "entropy",
        "laplacian_var",
        "foreground_bbox_area_ratio",
    ]
    summary: dict[str, Any] = {
        "files": len(paths),
        "ok": len(metadata),
        "errors": errors[:20],
        "mode_counts": counter_dict(Counter(r["mode"] for r in metadata)),
        "format_counts": counter_dict(Counter(r["format"] for r in metadata)),
        "dimension_counts": counter_dict(Counter((r["width"], r["height"]) for r in metadata)),
        "dpi_counts": counter_dict(Counter((r["dpi_x"], r["dpi_y"]) for r in metadata)),
        "stats": {
            key: describe([float(r[key]) for r in metadata if r.get(key) is not None])
            for key in metadata_numeric_keys
        },
        "quality_sample_files": len(quality_paths),
        "quality_sample_ok": len(quality_records),
        "quality_sample_errors": quality_errors[:20],
        "quality_stats": {
            key: describe([float(r[key]) for r in quality_records if r.get(key) is not None])
            for key in quality_numeric_keys
        },
        "samples": quality_records[:sample_records],
    }
    if hash_files:
        hashes = Counter(r["sha256"] for r in metadata if r.get("sha256"))
        summary["unique_sha256"] = len(hashes)
        summary["duplicate_sha256_groups"] = sum(1 for value in hashes.values() if value > 1)
        summary["duplicate_sha256_files"] = sum(value for value in hashes.values() if value > 1)

    anomaly_keys = [
        ("smallest_area", "area", False),
        ("largest_area", "area", True),
        ("lowest_std", "std", False),
        ("highest_white_frac", "white_frac_gte_250", True),
        ("lowest_entropy", "entropy", False),
        ("lowest_laplacian_var", "laplacian_var", False),
    ]
    for out_key, sort_key, reverse in anomaly_keys:
        source = metadata if sort_key in metadata_numeric_keys else quality_records
        summary[out_key] = [
            {
                "path": r["path"],
                "width": r["width"],
                "height": r["height"],
                "std": r.get("std"),
                "white_frac_gte_250": r.get("white_frac_gte_250"),
                "entropy": r.get("entropy"),
                "laplacian_var": r.get("laplacian_var"),
            }
            for r in sorted(source, key=lambda item: item[sort_key], reverse=reverse)[:10]
        ]
    return summary


def pngs(root: Path) -> list[Path]:
    if not root.exists():
        return []
    return sorted(root.rglob("*.png"))


def read_csv(path: Path) -> list[dict[str, str]]:
    with path.open(newline="", encoding="utf-8") as handle:
        return list(csv.DictReader(handle))


def subject_from_path(path: str) -> str | None:
    match = re.search(r"/(\d{8})/", path)
    return match.group(1) if match else None


def fgp_from_roll_name(path: str) -> str | None:
    match = re.search(r"_roll_(\d{2})\.png$", path)
    return match.group(1) if match else None


def device_ppi_from_nist302g(path: str) -> tuple[str | None, str | None, str | None]:
    parts = Path(path).parts
    for idx, part in enumerate(parts):
        if part in {"R", "S", "U", "V", "C"} and idx + 1 < len(parts):
            return part, parts[idx + 1], "challenger" if "challengers" in parts else "baseline"
    return None, None, None


def summarize_manifest(path: Path) -> dict[str, Any]:
    rows = read_csv(path)
    mates = [row.get("mate", "") for row in rows if row.get("mate")]
    unique_mates = sorted(set(mates))
    mate_counter = Counter(mates)
    fanouts = list(mate_counter.values())
    device_counts: Counter[tuple[str | None, str | None, str | None]] = Counter(
        device_ppi_from_nist302g(mate) for mate in unique_mates
    )
    row_device_counts: Counter[tuple[str | None, str | None, str | None]] = Counter(
        device_ppi_from_nist302g(mate) for mate in mates
    )
    return {
        "path": str(path),
        "rows": len(rows),
        "unique_mates": len(unique_mates),
        "unique_latents": len({row.get("latent", "") for row in rows if row.get("latent")}),
        "unique_identity_labels": len({row.get("identity_label", "") for row in rows if row.get("identity_label")}),
        "unique_subjects": len({row.get("subject", "") for row in rows if row.get("subject")}),
        "fgp_counts": counter_dict(Counter(row.get("fgp", "") for row in rows if row.get("fgp"))),
        "unique_mate_device_ppi_counts": counter_dict(device_counts),
        "row_mate_device_ppi_counts": counter_dict(row_device_counts),
        "latent_per_mate": describe([float(v) for v in fanouts]),
        "top_mate_fanout": [
            {"mate": mate, "rows": count}
            for mate, count in mate_counter.most_common(10)
        ],
    }


def collect_unique_manifest_mates(manifest_paths: list[Path]) -> list[Path]:
    mates: set[str] = set()
    for manifest_path in manifest_paths:
        for row in read_csv(manifest_path):
            mate = row.get("mate")
            if mate:
                mates.add(mate)
    return sorted(Path(mate) for mate in mates)


def summarize_nist302a_group(paths: list[Path]) -> dict[str, Any]:
    subjects = Counter(subject_from_path(str(path)) for path in paths)
    fingers = Counter(fgp_from_roll_name(path.name) for path in paths)
    per_subject = Counter(subject for subject in subjects if subject)
    return {
        "unique_subjects": len([subject for subject in subjects if subject]),
        "fgp_counts": counter_dict(fingers),
        "images_per_subject": describe([float(v) for k, v in per_subject.items() if k]),
    }


def main() -> int:
    parser = argparse.ArgumentParser()
    parser.add_argument("--repo-root", default=".")
    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_hq_image_groups.json")
    parser.add_argument("--sample-records", type=int, default=5)
    parser.add_argument("--quality-limit", type=int, default=500)
    parser.add_argument("--hash-files", action="store_true")
    args = parser.parse_args()

    repo_root = Path(args.repo_root).resolve()
    dataset_root = Path(args.dataset_root).resolve()
    manifest_root = Path(args.manifest_root).resolve()

    nist302g_root = repo_root / "data/converted/nist302g_png"
    nist302a_root = dataset_root / "nist302a/images/challengers"

    groups: dict[str, list[Path]] = {
        "nist302g_baseline_V_1000_roll": pngs(nist302g_root / "baseline/irr/V/1000"),
        "nist302g_baseline_U_1000_plain": pngs(nist302g_root / "baseline/irr/U/1000"),
        "nist302g_baseline_R_1000_slap": pngs(nist302g_root / "baseline/irr/R/1000"),
        "nist302g_baseline_S_500_slap": pngs(nist302g_root / "baseline/irr/S/500"),
        "nist302g_challenger_C_500_roll": pngs(nist302g_root / "challengers/irr/C/500"),
    }

    for challenger in "ABCDEFGH":
        groups[f"nist302a_challenger_{challenger}_500_roll"] = pngs(
            nist302a_root / challenger / "roll/png"
        )
    groups["nist302a_challengers_all_500_roll"] = sorted(
        path for challenger in "ABCDEFGH" for path in pngs(nist302a_root / challenger / "roll/png")
    )

    ready_manifests = sorted(manifest_root.glob("*_ready/paired_302i_*.csv"))
    ready_manifests = [path for path in ready_manifests if not path.name.endswith("_with_irr.csv")]
    manifest_mates = collect_unique_manifest_mates(ready_manifests)
    groups["nist302i_ready_unique_mates"] = manifest_mates

    report: dict[str, Any] = {
        "paths": {
            "repo_root": str(repo_root),
            "dataset_root": str(dataset_root),
            "manifest_root": str(manifest_root),
            "nist302g_converted_root": str(nist302g_root),
            "nist302a_challengers_root": str(nist302a_root),
        },
        "manifest_summaries": [summarize_manifest(path) for path in ready_manifests],
        "groups": {},
        "cross_coverage": {},
    }

    nist302g_sets = {name: {str(path) for path in paths} for name, paths in groups.items() if name.startswith("nist302g_")}
    manifest_mate_set = {str(path) for path in manifest_mates}
    report["cross_coverage"]["nist302i_ready_mates_in_nist302g_groups"] = {
        name: len(manifest_mate_set & path_set) for name, path_set in nist302g_sets.items()
    }

    for name, paths in groups.items():
        image_summary = summarize_images(paths, args.sample_records, args.hash_files, args.quality_limit)
        if name.startswith("nist302a_challenger"):
            image_summary["identity_summary"] = summarize_nist302a_group(paths)
        report["groups"][name] = image_summary
        print(json.dumps({"group": name, "files": len(paths), "ok": image_summary["ok"]}))

    out_path = Path(args.out)
    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), "groups": len(groups)}, indent=2))
    return 0


if __name__ == "__main__":
    raise SystemExit(main())