File size: 4,257 Bytes
1e55229 76c63a5 1e55229 b61b283 1e55229 b61b283 1e55229 b61b283 1e55229 76c63a5 1e55229 | 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 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 | #!/usr/bin/env python3
"""Export the Hugging Face VCR dataset back to the original r2c layout."""
from __future__ import annotations
import argparse
import json
import os
import shutil
from itertools import islice
from pathlib import Path
from datasets import load_dataset, load_from_disk
from huggingface_hub import hf_hub_download
from tqdm import tqdm
def main() -> None:
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument("repo_id", help="Hugging Face dataset repo id, e.g. rowanz/vcr")
parser.add_argument("output_dir", type=Path, help="Directory to create in r2c-compatible layout")
parser.add_argument("--token", default=None, help="HF token for private/gated repos")
parser.add_argument("--limit", type=int, default=None, help="Optional per-split row limit for smoke tests")
parser.add_argument("--local-stage-dir", type=Path, default=None, help="Use a local vcr_hf.py stage directory instead of the Hub")
args = parser.parse_args()
args.output_dir.mkdir(parents=True, exist_ok=True)
images_dir = args.output_dir / "vcr1images"
images_dir.mkdir(parents=True, exist_ok=True)
for split in ("train", "val", "test"):
if args.local_stage_dir is None:
src = hf_hub_download(
repo_id=args.repo_id,
repo_type="dataset",
filename=f"original_annotations/{split}.jsonl",
token=args.token,
)
else:
src = args.local_stage_dir / "original_annotations" / f"{split}.jsonl"
dst = args.output_dir / f"{split}.jsonl"
if args.limit is None:
shutil.copyfile(src, dst)
else:
with open(src, "r", encoding="utf-8") as f_in, open(dst, "w", encoding="utf-8") as f_out:
for i, line in enumerate(f_in):
if i >= args.limit:
break
f_out.write(line)
if args.local_stage_dir is None:
if args.limit is None:
split_expr = {s: s for s in ("train", "validation", "test")}
ds = load_dataset(args.repo_id, "image_examples", split=split_expr, token=args.token)
if not isinstance(ds, dict):
ds = {"train": ds}
else:
ds = {
split_name: load_dataset(
args.repo_id,
"image_examples",
split=split_name,
streaming=True,
token=args.token,
)
for split_name in ("train", "validation", "test")
}
else:
local_ds = load_from_disk(args.local_stage_dir / "image_examples.dataset")
ds = {}
for split_name in ("train", "validation", "test"):
ds[split_name] = local_ds[split_name].select(range(min(args.limit, len(local_ds[split_name])))) if args.limit else local_ds[split_name]
seen: set[str] = set()
for split_name, split_ds in ds.items():
rows = islice(split_ds, args.limit) if args.limit is not None else split_ds
for row in tqdm(rows, desc=f"export {split_name} images", total=args.limit):
img_fn = row["img_fn"]
metadata_fn = row["metadata_fn"]
if img_fn not in seen:
image = row["image"]
dst_img = images_dir / img_fn
dst_img.parent.mkdir(parents=True, exist_ok=True)
image.save(dst_img)
seen.add(img_fn)
if metadata_fn not in seen:
metadata = {
"boxes": row["boxes"],
"segms": json.loads(row["segms_json"]),
"names": row["objects"],
"width": row["width"],
"height": row["height"],
}
dst_meta = images_dir / metadata_fn
dst_meta.parent.mkdir(parents=True, exist_ok=True)
with open(dst_meta, "w", encoding="utf-8") as f:
json.dump(metadata, f, separators=(",", ":"))
seen.add(metadata_fn)
print(f"Wrote r2c-compatible VCR layout to {args.output_dir}", flush=True)
os._exit(0)
if __name__ == "__main__":
main()
|