File size: 5,513 Bytes
47709ed | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 | #!/usr/bin/env python3
"""Prepare deterministic train/validation relation pools from official CLEVR."""
from __future__ import annotations
import argparse
import hashlib
import json
import random
import zipfile
from collections import Counter
from datetime import datetime, timezone
from pathlib import Path
def parse_args() -> argparse.Namespace:
parser = argparse.ArgumentParser()
parser.add_argument(
"--archive",
type=Path,
default=(Path(__file__).resolve().parents[2] / "datasets/raw/clevr/CLEVR_v1.0.zip"),
)
parser.add_argument(
"--output-dir",
type=Path,
default=Path("./datasets/clevr_relations"),
)
parser.add_argument("--train-pool", type=int, default=1024)
parser.add_argument("--validation-pool", type=int, default=256)
parser.add_argument("--seed", type=int, default=20260829)
return parser.parse_args()
def sha256_file(path: Path) -> str:
digest = hashlib.sha256()
with path.open("rb") as handle:
for chunk in iter(lambda: handle.read(16 << 20), b""):
digest.update(chunk)
return digest.hexdigest()
def read_questions(archive: zipfile.ZipFile, split: str) -> list[dict]:
member = f"CLEVR_v1.0/questions/CLEVR_{split}_questions.json"
with archive.open(member) as handle:
payload = json.load(handle)
return payload["questions"]
def has_relation_program(sample: dict) -> bool:
return any(step.get("function") == "relate" for step in sample.get("program", []))
def choose_pool(rows: list[dict], count: int, seed: int) -> list[dict]:
eligible = [row for row in rows if has_relation_program(row)]
order = list(range(len(eligible)))
random.Random(seed).shuffle(order)
return [eligible[index] for index in order[: min(count, len(order))]]
def materialize(
archive: zipfile.ZipFile,
rows: list[dict],
split: str,
output_dir: Path,
) -> list[dict]:
image_dir = output_dir / "images" / split
image_dir.mkdir(parents=True, exist_ok=True)
prepared = []
for row in rows:
filename = row["image_filename"]
target = image_dir / filename
if not target.exists():
member = f"CLEVR_v1.0/images/{split}/{filename}"
with archive.open(member) as source, target.open("wb") as sink:
while chunk := source.read(1 << 20):
sink.write(chunk)
functions = [step.get("function") for step in row.get("program", [])]
prepared.append(
{
"split": split,
"question_index": int(row["question_index"]),
"image_index": int(row["image_index"]),
"question_family_index": int(row["question_family_index"]),
"question": row["question"],
"answer": str(row["answer"]),
"image_filename": filename,
"image_path": str(target),
"program_functions": functions,
"relation_count": functions.count("relate"),
}
)
return prepared
def summarize(rows: list[dict]) -> dict:
return {
"rows": len(rows),
"unique_images": len({row["image_index"] for row in rows}),
"answer_distribution": dict(Counter(row["answer"] for row in rows)),
"relation_count_distribution": dict(
Counter(str(row["relation_count"]) for row in rows)
),
"question_family_count": len(
{row["question_family_index"] for row in rows}
),
}
def main() -> None:
args = parse_args()
if not args.archive.exists():
raise FileNotFoundError(args.archive)
args.output_dir.mkdir(parents=True, exist_ok=True)
with zipfile.ZipFile(args.archive) as archive:
train_all = read_questions(archive, "train")
validation_all = read_questions(archive, "val")
train_selected = choose_pool(train_all, args.train_pool, args.seed)
validation_selected = choose_pool(
validation_all, args.validation_pool, args.seed + 1
)
train_rows = materialize(archive, train_selected, "train", args.output_dir)
validation_rows = materialize(
archive, validation_selected, "val", args.output_dir
)
train_path = args.output_dir / "train_relation_pool.json"
validation_path = args.output_dir / "validation_relation_pool.json"
train_path.write_text(
json.dumps(train_rows, ensure_ascii=False, indent=2), encoding="utf-8"
)
validation_path.write_text(
json.dumps(validation_rows, ensure_ascii=False, indent=2), encoding="utf-8"
)
manifest = {
"dataset": "CLEVR v1.0",
"source": "https://dl.fbaipublicfiles.com/clevr/CLEVR_v1.0.zip",
"license": "CC BY 4.0",
"archive": str(args.archive),
"archive_bytes": args.archive.stat().st_size,
"archive_sha256": sha256_file(args.archive),
"selection": "program contains at least one relate operation",
"seed": args.seed,
"train": summarize(train_rows),
"validation": summarize(validation_rows),
"test_rows_read": 0,
"prepared_at": datetime.now(timezone.utc).isoformat(),
}
manifest_path = args.output_dir / "dataset_manifest.json"
manifest_path.write_text(
json.dumps(manifest, ensure_ascii=False, indent=2), encoding="utf-8"
)
print(json.dumps(manifest, ensure_ascii=False, indent=2))
if __name__ == "__main__":
main()
|