download_hf_complement.py

#!/usr/bin/env python3
"""Download the Hugging Face complement for the current EPIC-KITCHENS tree.

Run from the dataset root:

    python3 download_hf_complement.py

This script:
1. Lists all video files on the Hugging Face dataset mirror.
2. Compares them against the local `P*/videos/*.MP4` files.
3. Writes the HF-only complement manifest to `annotations/hf_complement_paths.txt`.
4. Downloads only those missing files into the current directory layout.
"""

import argparse
import json
import os
from pathlib import Path
import re
import sys
import urllib.request


HF_DATASET_ID = "a1raman/epic_kitchens_100"
HF_TREE_URL = (
    "https://huggingface.co/api/datasets/"
    f"{HF_DATASET_ID}/tree/main?recursive=true&expand=false"
)
VIDEO_PATH_RE = re.compile(r"^P\d{2}/videos/P\d{2}_\d+\.MP4$")


def parse_args() -> argparse.Namespace:
    parser = argparse.ArgumentParser(
        description="Download the HF-only complement into the current dataset tree."
    )
    parser.add_argument(
        "--root",
        type=Path,
        default=Path("."),
        help="Dataset root directory. Default: current directory.",
    )
    parser.add_argument(
        "--manifest",
        type=Path,
        default=Path("annotations/hf_complement_paths.txt"),
        help="Relative path of the generated manifest file.",
    )
    parser.add_argument(
        "--repo-id",
        default=HF_DATASET_ID,
        help=f"Hugging Face dataset repo id. Default: {HF_DATASET_ID}",
    )
    parser.add_argument(
        "--revision",
        default="main",
        help="Repo revision to download from. Default: main.",
    )
    parser.add_argument(
        "--max-workers",
        type=int,
        default=8,
        help="Concurrent download workers for snapshot_download. Default: 8.",
    )
    parser.add_argument(
        "--manifest-only",
        action="store_true",
        help="Only write the complement manifest and skip downloading.",
    )
    parser.add_argument(
        "--no-hf-transfer",
        action="store_true",
        help="Do not enable HF_HUB_ENABLE_HF_TRANSFER=1.",
    )
    return parser.parse_args()


def fetch_remote_video_paths():
    with urllib.request.urlopen(HF_TREE_URL, timeout=120) as response:
        data = json.load(response)
    return sorted(
        item["path"]
        for item in data
        if item.get("type") == "file" and VIDEO_PATH_RE.match(item.get("path", ""))
    )


def collect_local_video_paths(root: Path):
    return {path.as_posix() for path in root.glob("P*/videos/*.MP4")}


def main() -> int:
    args = parse_args()
    root = args.root.resolve()
    manifest_path = (root / args.manifest).resolve()
    manifest_path.parent.mkdir(parents=True, exist_ok=True)

    remote_paths = fetch_remote_video_paths()
    local_paths = collect_local_video_paths(root)
    complement_paths = [path for path in remote_paths if path not in local_paths]

    manifest_path.write_text("\n".join(complement_paths) + ("\n" if complement_paths else ""))
    print(f"Wrote {len(complement_paths)} paths to {manifest_path}")

    if args.manifest_only:
        return 0

    if not complement_paths:
        print("No HF-only complement files to download.")
        return 0

    if not args.no_hf_transfer:
        # This variable must be set before importing huggingface_hub.
        os.environ["HF_HUB_ENABLE_HF_TRANSFER"] = "1"

    try:
        from huggingface_hub import snapshot_download
    except ImportError:
        print(
            "Missing dependency: install with "
            '`python3 -m pip install -U "huggingface_hub[hf_transfer]"`',
            file=sys.stderr,
        )
        return 2

    snapshot_download(
        repo_id=args.repo_id,
        repo_type="dataset",
        revision=args.revision,
        local_dir=str(root),
        allow_patterns=complement_paths,
        max_workers=args.max_workers,
    )
    print(f"Downloaded {len(complement_paths)} files into {root}")
    return 0


if __name__ == "__main__":
    raise SystemExit(main())