File size: 7,050 Bytes
6f003f2 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 | #!/usr/bin/env python
"""Validate the fixed-revision RECAP LIBERO-10 Task 0 local dataset.
The expected video counts come from the file manifest at the pinned revision,
not from the historical SFT ``info.json`` ``total_videos`` field.
"""
from __future__ import annotations
import argparse
import json
import math
import sys
from pathlib import Path
from typing import Any
import numpy as np
import pyarrow.parquet as pq
SCRIPT_DIR = Path(__file__).resolve().parent
PROJECT_ROOT = SCRIPT_DIR.parent
if str(SCRIPT_DIR) not in sys.path:
sys.path.insert(0, str(SCRIPT_DIR))
from task0_baseline_common import require_project_path # noqa: E402
REVISION = "75b382d2c066bcedd8b030285b45c856913f1497"
SPLITS: dict[str, dict[str, int]] = {
"libero10_task0_sft": {"episodes": 30, "frames": 8005, "image": 30, "wrist_image": 30},
"libero10_task0_train": {"episodes": 4096, "frames": 1561874, "image": 4096, "wrist_image": 4096},
"libero10_task0_eval": {"episodes": 64, "frames": 25493, "image": 64, "wrist_image": 64},
}
def _check_video_files(split_path: Path, expected: dict[str, int]) -> dict[str, Any]:
report: dict[str, Any] = {}
for key in ("image", "wrist_image"):
files = sorted((split_path / "videos").glob(f"**/{key}/episode_*.mp4"))
names = {item.name for item in files}
expected_names = {f"episode_{index:06d}.mp4" for index in range(expected["episodes"])}
report[key] = {
"count": len(files),
"expected_count": expected[key],
"missing_names": sorted(expected_names - names)[:20],
"unexpected_names": sorted(names - expected_names)[:20],
"bytes": sum(item.stat().st_size for item in files),
}
return report
def _check_sidecar(sidecar: Path, expected_rows: int) -> dict[str, Any]:
table = pq.read_table(sidecar, columns=["episode_index", "frame_index", "return", "reward"])
episode = table.column("episode_index").to_numpy()
frame = table.column("frame_index").to_numpy()
returns = table.column("return").to_numpy()
rewards = table.column("reward").to_numpy()
keys = np.stack([episode, frame], axis=1)
return {
"rows": table.num_rows,
"expected_rows": expected_rows,
"unique_keys": int(np.unique(keys, axis=0).shape[0]),
"finite": bool(np.isfinite(returns).all() and np.isfinite(rewards).all()),
"return_min": float(returns.min()) if len(returns) else math.nan,
"return_max": float(returns.max()) if len(returns) else math.nan,
}
def _decode_samples(split_path: Path, expected: dict[str, int]) -> list[str]:
try:
import av
except Exception as exc: # pragma: no cover - environment diagnostic
return [f"PyAV import failed: {type(exc).__name__}: {exc}"]
errors: list[str] = []
sample_indices = sorted({0, expected["episodes"] // 2, expected["episodes"] - 1})
for key in ("image", "wrist_image"):
for index in sample_indices:
chunk = index // 1000
path = split_path / "videos" / f"chunk-{chunk:03d}" / key / f"episode_{index:06d}.mp4"
if not path.is_file():
errors.append(f"missing sample: {path}")
continue
try:
with av.open(str(path)) as container:
frame = next(container.decode(video=0), None)
if frame is None:
errors.append(f"no decodable frame: {path}")
except Exception as exc:
errors.append(f"decode failed {path}: {type(exc).__name__}: {exc}")
return errors
def validate(data_root: Path, *, returns_tag: str | None = None, decode_samples: bool = False) -> dict[str, Any]:
report: dict[str, Any] = {"data_root": str(data_root), "revision": REVISION, "splits": {}, "errors": []}
for split, expected in SPLITS.items():
path = data_root / split
split_report: dict[str, Any] = {"path": str(path), "expected": expected}
info_path = path / "meta" / "info.json"
if not info_path.is_file():
report["errors"].append(f"missing {info_path}")
continue
info = json.loads(info_path.read_text(encoding="utf-8"))
parquet_files = sorted((path / "data").rglob("*.parquet"))
rows = sum(int(pq.ParquetFile(item).metadata.num_rows) for item in parquet_files)
split_report.update(
{
"metadata_episodes": info.get("total_episodes"),
"metadata_frames": info.get("total_frames"),
"metadata_total_videos": info.get("total_videos"),
"parquet_files": len(parquet_files),
"expected_parquet_files": expected["episodes"],
"parquet_rows": rows,
"expected_rows": expected["frames"],
"videos": _check_video_files(path, expected),
}
)
if len(parquet_files) != expected["episodes"]:
report["errors"].append(f"{split}: parquet file count mismatch")
if rows != expected["frames"]:
report["errors"].append(f"{split}: parquet row count mismatch")
for key, video_report in split_report["videos"].items():
if video_report["count"] != expected[key] or video_report["missing_names"] or video_report["unexpected_names"]:
report["errors"].append(f"{split}: {key} video manifest mismatch")
if decode_samples:
split_report["decode_sample_errors"] = _decode_samples(path, expected)
report["errors"].extend(f"{split}: {item}" for item in split_report["decode_sample_errors"])
if returns_tag:
sidecar = path / "meta" / f"returns_{returns_tag}.parquet"
if not sidecar.is_file():
report["errors"].append(f"{split}: missing {sidecar}")
else:
split_report["returns"] = _check_sidecar(sidecar, expected["frames"])
if (
split_report["returns"]["rows"] != expected["frames"]
or split_report["returns"]["unique_keys"] != expected["frames"]
or not split_report["returns"]["finite"]
):
report["errors"].append(f"{split}: returns sidecar mismatch")
report["splits"][split] = split_report
report["ok"] = not report["errors"]
return report
def main() -> int:
parser = argparse.ArgumentParser()
parser.add_argument("--data-root", type=Path, default=Path("datasets/recap_libero10_task0"))
parser.add_argument("--returns-tag", default=None)
parser.add_argument("--decode-samples", action="store_true")
args = parser.parse_args()
root = require_project_path(args.data_root, label="data root")
report = validate(root, returns_tag=args.returns_tag, decode_samples=args.decode_samples)
print(json.dumps(report, ensure_ascii=False, indent=2, sort_keys=True))
return 0 if report["ok"] else 1
if __name__ == "__main__":
raise SystemExit(main())
|