code/adapter_drive.py

#!/usr/bin/env python3
"""
DRIVE adapter — vessel segmentation public dataset.

Inputs:
    {input_root}/DRIVE/training/images/NN_training.tif       (20 files)
    {input_root}/DRIVE/training/1st_manual/NN_manual1.gif    (vessel GT)
    {input_root}/DRIVE/training/mask/NN_training_mask.gif    (FOV mask)

Outputs (under {output_root}):
    extracted/public_drive_vessel/{hash[:2]}/{hash}/
        fundus_color.jpg     (re-encoded from .tif)
        vessel_mask.png      (binary 0/255, from 1st_manual)
        fov_mask.png         (binary 0/255, from mask/)
        meta.json
    manifest/public_drive_vessel_images.parquet
    captions/public_drive_vessel_captions.parquet

Test split (also 20 images) intentionally skipped — it lacks vessel GT, only FOV mask.
"""
import argparse
import json
from pathlib import Path

import numpy as np
import pandas as pd
from PIL import Image

from public_common import (
    IMAGE_SCHEMA_COLUMNS, CAPTION_SCHEMA_COLUMNS,
    study_hash_for, default_base_fields,
    caption_l1_public, caption_l3_public,
    study_dir_for, rel_file_path, write_meta, coerce_image_row,
)

COHORT = "public_drive_vessel"
COHORT_PHRASE = "DRIVE retinal vessel segmentation dataset"


def _binarize(p: Path) -> np.ndarray:
    arr = np.array(Image.open(p).convert("L"))
    return ((arr > 127).astype(np.uint8) * 255)


def process_one(image_path: Path, vessel_path: Path, fov_path: Path,
                out_root: Path, force: bool):
    basename = image_path.stem  # "21_training"
    sh = study_hash_for(COHORT, basename)
    sdir = study_dir_for(out_root, COHORT, sh)
    sdir.mkdir(parents=True, exist_ok=True)

    meta_path = sdir / "meta.json"
    if meta_path.exists() and not force:
        try:
            meta = json.loads(meta_path.read_text())
            if meta.get("status") == "ok":
                return _row_and_caps(meta)
        except Exception:
            pass

    img = Image.open(image_path).convert("RGB")
    w, h = img.size
    img.save(sdir / "fundus_color.jpg", "JPEG", quality=95)

    Image.fromarray(_binarize(vessel_path), mode="L").save(
        sdir / "vessel_mask.png", "PNG", optimize=True)
    Image.fromarray(_binarize(fov_path), mode="L").save(
        sdir / "fov_mask.png", "PNG", optimize=True)

    meta = {
        "status": "ok",
        "cohort": COHORT,
        "study_hash": sh,
        "source_basename": basename,
        "image_height_px": int(h),
        "image_width_px": int(w),
        "has_vessel_mask": True,
        "has_fov_mask": True,
        "eye": "unknown",
    }
    write_meta(sdir, meta)
    return _row_and_caps(meta)


def _row_and_caps(meta: dict):
    sh = meta["study_hash"]
    image_id = f"{COHORT}_{sh}_fundus_color"
    row = default_base_fields(COHORT, sh, eye=meta["eye"])
    row.update({
        "image_id": image_id,
        "file_path": rel_file_path(COHORT, sh, "fundus_color.jpg"),
        "file_format": "jpg",
        "modality": "fundus_color",
        "anatomy": "macula",
        "device_technology": "fundus_camera",
        "scan_protocol": "single_shot",
        "image_height_px": int(meta["image_height_px"]),
        "image_width_px": int(meta["image_width_px"]),
        "has_segmentation": True,
        "n_layers_visible": 0,
        "is_valid": True,
    })
    caps = caption_l1_public(image_id, COHORT_PHRASE, "fundus_color", meta["eye"])
    l3 = ("A color fundus photograph from the DRIVE dataset, "
          "with manually annotated retinal vessel segmentation mask and field-of-view mask.")
    caps.append(caption_l3_public(image_id, l3, "manifest_fields+vessel_mask"))
    return row, caps


def main():
    ap = argparse.ArgumentParser()
    ap.add_argument("--input-root", required=True,
                    help="Path to .../Generation/DRIVE (contains DRIVE/training/...)")
    ap.add_argument("--output-root", required=True,
                    help="Public manifest root, e.g. .../c3/public_eye_pretrain")
    ap.add_argument("--force", action="store_true")
    args = ap.parse_args()

    in_root = Path(args.input_root)
    out_root = Path(args.output_root)
    train_dir = in_root / "DRIVE" / "training"
    img_dir = train_dir / "images"
    vessel_dir = train_dir / "1st_manual"
    fov_dir = train_dir / "mask"

    img_files = sorted(img_dir.glob("*_training.tif"))
    print(f"[{COHORT}] {len(img_files)} training images under {img_dir}")

    rows, caps = [], []
    for ip in img_files:
        stem = ip.stem.replace("_training", "")
        vp = vessel_dir / f"{stem}_manual1.gif"
        fp = fov_dir / f"{stem}_training_mask.gif"
        if not (vp.exists() and fp.exists()):
            print(f"  skip {ip.name}: vessel={vp.exists()}, fov={fp.exists()}")
            continue
        row, cap = process_one(ip, vp, fp, out_root, args.force)
        rows.append(row)
        caps.extend(cap)

    if not rows:
        print(f"[{COHORT}] no rows produced, aborting")
        return

    manifest_dir = out_root / "manifest"
    captions_dir = out_root / "captions"
    manifest_dir.mkdir(parents=True, exist_ok=True)
    captions_dir.mkdir(parents=True, exist_ok=True)

    imgs_df = pd.DataFrame([coerce_image_row(r) for r in rows])[IMAGE_SCHEMA_COLUMNS]
    imgs_df.to_parquet(manifest_dir / f"{COHORT}_images.parquet", index=False)

    caps_df = pd.DataFrame(caps)[CAPTION_SCHEMA_COLUMNS]
    caps_df.to_parquet(captions_dir / f"{COHORT}_captions.parquet", index=False)

    print(f"[{COHORT}] wrote {len(imgs_df)} image rows, {len(caps_df)} caption rows")


if __name__ == "__main__":
    main()