File size: 3,518 Bytes
ab7f3d0 9ceda63 ab7f3d0 9ceda63 ab7f3d0 | 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 | #!/usr/bin/env python3
"""Load every unified action22/state23 episode using unmodified official LeRobot."""
import argparse
import json
import os
from pathlib import Path
def collection_root(root, name="all"):
# Historical CLI alias selects the same full set; it never creates a subset.
if name == "main":
name = "all"
root = Path(root).resolve()
registry = json.loads((root / "meta/collections.json").read_text())
if registry.get("default_collection") != "all" or [r["name"] for r in registry["collections"]] != ["all"]:
raise ValueError("This loader requires the single unified all collection")
row = next((r for r in registry["collections"] if r["name"] == name), None)
if row is None:
raise ValueError(f"Unknown collection {name}; available: {[r['name'] for r in registry['collections']]}")
rel = Path(row["root"])
if rel.is_absolute() or ".." in rel.parts:
raise ValueError("Collection root must stay inside the extracted dataset")
target = (root / rel).resolve()
if not target.is_relative_to(root):
raise ValueError("Collection root escapes dataset")
return target
def load_collection(root, name="all", episodes=None, cache_dir=None):
os.environ["HF_HUB_OFFLINE"] = "1"
os.environ["HF_DATASETS_OFFLINE"] = "1"
os.environ["TOKENIZERS_PARALLELISM"] = "false"
if cache_dir is not None:
os.environ["HF_DATASETS_CACHE"] = str(Path(cache_dir).resolve())
from lerobot.datasets.lerobot_dataset import LeRobotDataset
dataset = LeRobotDataset(
repo_id=f"local/robotagentbench-{name}", root=collection_root(root, name), episodes=episodes,
download_videos=False, video_backend="pyav", tolerance_s=0.0003,
)
# Specialize only in-memory metadata for a uniformly sized selection.
dimensions = collection_root(root, name) / "meta/video_dimensions.jsonl"
if dimensions.exists() and episodes is not None:
selected = set(episodes)
shapes = {}
for line in dimensions.open():
row = json.loads(line)
if row["episode_index"] in selected:
shapes.setdefault(row["video_key"], set()).add((row["height"], row["width"], 3))
for key, sizes in shapes.items():
if len(sizes) == 1:
shape = next(iter(sizes))
dataset.meta.info["features"][key]["shape"] = shape
dataset.meta.info["features"][key]["video_info"].update({"video.height": shape[0], "video.width": shape[1]})
return dataset
if __name__ == "__main__":
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument("--root", type=Path, required=True, help="Extracted dataset directory containing meta/collections.json")
parser.add_argument("--collection", default="all")
parser.add_argument("--episode", type=int, default=0)
parser.add_argument("--frame", type=int, default=0)
parser.add_argument("--cache-dir", type=Path)
args = parser.parse_args()
dataset = load_collection(args.root, args.collection, [args.episode], args.cache_dir)
sample = dataset[args.frame]
print(json.dumps({"root": str(collection_root(args.root, args.collection)), "collection": args.collection,
"selected_episode": args.episode, "selected_episode_frames": len(dataset),
"task": sample.get("task"),
"sample_shapes": {k: list(v.shape) for k, v in sample.items() if hasattr(v, "shape")}}, indent=2))
|