Datasets:
File size: 1,311 Bytes
292d868 | 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 | """Load a downloaded ImageFolder snapshot and check its full ID population."""
import argparse
import json
from pathlib import Path
from datasets import Image, load_dataset
def load(root):
dataset = load_dataset("imagefolder", data_dir=str(root / "data"),
split="test", drop_labels=True)
expected = {json.loads(line)["image_id"] for line in
(root / "release/image_manifest.jsonl").read_text().splitlines()}
assert len(dataset) == 11534
assert set(dataset["image_id"]) == expected
assert "label" not in dataset.column_names, "Shard directory names are not labels"
row = dataset[0]
assert row["image"].size == (row["width"], row["height"])
streamed = load_dataset("imagefolder", data_dir=str(root / "data"),
split="test", drop_labels=True, streaming=True)
streamed = streamed.cast_column("image", Image(decode=False))
assert {row["image_id"] for row in streamed} == expected
print(f"Loaded {len(dataset):,} images; full streaming ID population matches.")
return dataset
if __name__ == "__main__":
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument("--root", type=Path, default=Path(__file__).resolve().parents[1])
load(parser.parse_args().root.resolve())
|