Datasets:
File size: 1,461 Bytes
1bd3e8e | 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 | #!/usr/bin/env python3
import argparse
import json
import tarfile
from pathlib import Path
parser = argparse.ArgumentParser()
parser.add_argument("--repo-root", type=Path, default=Path(__file__).resolve().parents[1])
parser.add_argument("--annotation", type=Path, required=True)
parser.add_argument("--output-root", type=Path, required=True)
args = parser.parse_args()
rows_by_tar = {}
with args.annotation.open("r", encoding="utf-8") as handle:
for line in handle:
if not line.strip():
continue
row = json.loads(line)
rows_by_tar.setdefault(row["metadata"]["tar"], []).append(row)
written = 0
for relative_tar, rows in sorted(rows_by_tar.items()):
with tarfile.open(args.repo_root / relative_tar) as archive:
for row in rows:
for output_key, member_key in (("source_views", "source_member"), ("target_view", "target_member")):
relative_output = row[output_key][0] if output_key == "source_views" else row[output_key]
output = args.output_root / relative_output
output.parent.mkdir(parents=True, exist_ok=True)
extracted = archive.extractfile(row["metadata"][member_key])
if extracted is None:
raise FileNotFoundError(row["metadata"][member_key])
output.write_bytes(extracted.read())
written += 1
print(f"materialized {written} images under {args.output_root}")
|