zju-eye-pretrain / code /adapter_eyepacs.py
MaybeRichard's picture
Initial upload: public_fundus (198k images, 42 shards) + manifest + captions + code
e2f75d1 verified
Raw
History Blame Contribute Delete
6.81 kB
#!/usr/bin/env python3
"""
EyePACS adapter — augmented_resized_V2 (143k DR-graded 600x600 fundus images).
Inputs:
{input_root}/augmented_resized_V2/{train,val,test}/{0,1,2,3,4}/*.jpg
Subdir name = DR grade. Filenames like 005b95c28852-600.jpg.
Outputs (under {output_root}):
extracted/public_eyepacs_combo_dr_aug/{hash[:2]}/{hash}/
fundus_color.jpg (copied as-is — already 600x600)
meta.json
manifest/public_eyepacs_combo_dr_aug_images.parquet
captions/public_eyepacs_combo_dr_aug_captions.parquet
143k images → multiprocessed. study_hash basename includes split to avoid hash collisions
if the same id_code appears across splits (it should not, but be defensive).
"""
import argparse
import json
import shutil
from concurrent.futures import ProcessPoolExecutor, as_completed
from pathlib import Path
import pandas as pd
from PIL import Image, ImageFile
ImageFile.LOAD_TRUNCATED_IMAGES = True
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_eyepacs_combo_dr_aug"
COHORT_PHRASE = "EyePACS combined diabetic retinopathy screening dataset (augmented 600x600 v2)"
DR_SEVERITY = {0: "none", 1: "mild", 2: "moderate", 3: "severe", 4: "proliferative"}
def process_one(args):
image_path_str, split, dr_grade, out_root_str, force = args
image_path = Path(image_path_str)
out_root = Path(out_root_str)
basename = f"{split}_{image_path.stem}" # e.g. train_005b95c28852-600
sh = study_hash_for(COHORT, basename)
sdir = study_dir_for(out_root, COHORT, sh)
sdir.mkdir(parents=True, exist_ok=True)
meta_p = sdir / "meta.json"
if meta_p.exists() and not force:
try:
meta = json.loads(meta_p.read_text())
if meta.get("status") == "ok":
return _row_and_caps(meta)
except Exception:
pass
dst_img = sdir / "fundus_color.jpg"
try:
if not dst_img.exists() or force:
shutil.copyfile(image_path, dst_img)
# Verify the copy is a readable image (catches truncated / non-JPG bytes)
with Image.open(dst_img) as im:
im.verify()
with Image.open(dst_img) as im:
w, h = im.size
except Exception as e:
# Delete the partial copy so a future run can re-attempt
try: dst_img.unlink(missing_ok=True)
except Exception: pass
return ("FAIL", basename, type(e).__name__, str(e)[:200])
meta = {
"status": "ok", "cohort": COHORT, "study_hash": sh,
"source_basename": basename, "split": split,
"image_height_px": int(h), "image_width_px": int(w),
"eye": "unknown",
"dr_grade": int(dr_grade),
}
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"])
dr = meta["dr_grade"]
row["diagnosis_group"] = ["DR"] if dr > 0 else []
row["severity"] = DR_SEVERITY[dr]
row["diagnosis_source"] = "screening_label"
row["label_confidence"] = "single_reader"
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": meta["image_height_px"],
"image_width_px": meta["image_width_px"],
"has_segmentation": False, "n_layers_visible": 0,
"is_valid": True,
})
caps = caption_l1_public(image_id, COHORT_PHRASE, "fundus_color", meta["eye"])
l3 = (f"A color fundus photograph from the EyePACS combined DR screening dataset "
f"({meta['split']} split, augmented 600x600), "
f"diabetic retinopathy grade {dr} ({DR_SEVERITY[dr]}).")
caps.append(caption_l3_public(image_id, l3, "manifest_fields+screening_label"))
return row, caps
def _enumerate_inputs(in_root: Path):
base = in_root / "augmented_resized_V2"
for split in ("train", "val", "test"):
for grade in range(5):
d = base / split / str(grade)
if not d.exists():
continue
for ip in d.glob("*.jpg"):
yield (str(ip), split, grade)
def main():
ap = argparse.ArgumentParser()
ap.add_argument("--input-root", required=True,
help="Path to .../Generation/EyePACS (contains augmented_resized_V2/)")
ap.add_argument("--output-root", required=True)
ap.add_argument("--num-workers", type=int, default=8)
ap.add_argument("--force", action="store_true")
ap.add_argument("--limit", type=int, default=None)
args = ap.parse_args()
in_root = Path(args.input_root)
out_root = Path(args.output_root)
inputs = list(_enumerate_inputs(in_root))
print(f"[{COHORT}] enumerated {len(inputs)} images")
if args.limit:
inputs = inputs[:args.limit]
job_args = [(p, s, g, str(out_root), args.force) for (p, s, g) in inputs]
rows, caps = [], []
failures = []
with ProcessPoolExecutor(max_workers=args.num_workers) as ex:
futs = [ex.submit(process_one, a) for a in job_args]
for i, fut in enumerate(as_completed(futs), 1):
try:
result = fut.result()
except Exception as e:
failures.append(("worker", type(e).__name__, str(e)[:200]))
continue
if isinstance(result, tuple) and len(result) == 4 and result[0] == "FAIL":
failures.append(result[1:])
continue
row, cap = result
rows.append(row)
caps.extend(cap)
if i % 5000 == 0:
print(f" ... {i}/{len(futs)} ({len(failures)} failed so far)")
if failures:
print(f"[{COHORT}] {len(failures)} images FAILED to extract:")
for f in failures[:30]:
print(f" {f}")
mdir = out_root / "manifest"; cdir = out_root / "captions"
mdir.mkdir(parents=True, exist_ok=True); cdir.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(mdir / f"{COHORT}_images.parquet", index=False)
caps_df = pd.DataFrame(caps)[CAPTION_SCHEMA_COLUMNS]
caps_df.to_parquet(cdir / f"{COHORT}_captions.parquet", index=False)
print(f"[{COHORT}] wrote {len(imgs_df)} images, {len(caps_df)} captions")
print(imgs_df.groupby(["severity"]).size().to_string())
if __name__ == "__main__":
main()