File size: 6,439 Bytes
adb8450 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 | # Copyright (c) Meta Platforms, Inc. and affiliates.
# All rights reserved.
#
# This source code is licensed under the license found in the
# LICENSE file in the root directory of this source tree.
"""Rebuild the audio-visual clips that the benchmark sessions refer to.
The dataset stores AudioSet clip IDs ("Y" + YouTube ID), not media. This script
looks up each clip's start time in the AudioSet segment CSVs, downloads the
ten-second segment from YouTube, and writes
<out-dir>/<asset_id>/frames/frame_00.jpg ... frame_09.jpg
<out-dir>/<asset_id>/audio.wav
Frames are spread evenly from the first to the last frame of the clip. Clips are
shared between users, so each clip is stored once. Clips that can no longer be
downloaded are listed in <out-dir>/missing.txt.
Requires yt-dlp and ffmpeg/ffprobe on PATH.
Usage:
python scripts/fetch_clips.py --data-dir data --out-dir media \\
--segments eval_segments.csv balanced_train_segments.csv unbalanced_train_segments.csv
"""
from __future__ import annotations
import argparse
import csv
import json
import subprocess
import sys
import tempfile
from pathlib import Path
FRAME_COUNT = 10
def load_segments(paths: list[Path]) -> dict[str, tuple[float, float]]:
"""YouTube ID -> (start, end) in seconds, from the AudioSet segment CSVs."""
segments: dict[str, tuple[float, float]] = {}
for path in paths:
with path.open(newline="") as handle:
for row in csv.reader(handle, skipinitialspace=True):
if not row or row[0].startswith("#"):
continue
segments[row[0]] = (float(row[1]), float(row[2]))
return segments
def session_assets(data_dir: Path, users: set[str] | None) -> set[str]:
"""Every clip ID used by a session of the selected users."""
assets: set[str] = set()
for path in sorted(data_dir.glob("*/7.1_dialogue_enhanced.json")):
user_id = path.parent.name
if users and user_id not in users:
continue
record = json.loads(path.read_text())
record = record[0] if isinstance(record, list) else record
assets.update(session["selected_asset_id"] for session in record["sessions"])
return assets
def youtube_id(asset_id: str) -> str:
return asset_id[1:] if len(asset_id) == 12 and asset_id.startswith("Y") else asset_id
def download_segment(video_id: str, start: float, end: float, target: Path) -> None:
subprocess.run(
[
"yt-dlp",
"--quiet",
"--no-warnings",
"--download-sections",
f"*{start}-{end}",
"--force-keyframes-at-cuts",
"-f",
"bv*+ba/b",
"--merge-output-format",
"mp4",
"-o",
str(target),
f"https://www.youtube.com/watch?v={video_id}",
],
check=True,
)
def duration(video: Path) -> float:
result = subprocess.run(
[
"ffprobe",
"-v",
"error",
"-show_entries",
"format=duration",
"-of",
"default=noprint_wrappers=1:nokey=1",
str(video),
],
check=True,
capture_output=True,
text=True,
)
return float(result.stdout.strip())
def write_media(video: Path, clip_dir: Path) -> None:
frames_dir = clip_dir / "frames"
frames_dir.mkdir(parents=True, exist_ok=True)
# Stop a little short of the end so the last seek still lands on a frame.
last = max(duration(video) - 0.05, 0.0)
for index in range(FRAME_COUNT):
moment = index * last / (FRAME_COUNT - 1)
subprocess.run(
[
"ffmpeg",
"-y",
"-loglevel",
"error",
"-ss",
f"{moment:.3f}",
"-i",
str(video),
"-frames:v",
"1",
"-q:v",
"2",
str(frames_dir / f"frame_{index:02d}.jpg"),
],
check=True,
)
subprocess.run(
[
"ffmpeg",
"-y",
"-loglevel",
"error",
"-i",
str(video),
"-vn",
"-acodec",
"pcm_s16le",
str(clip_dir / "audio.wav"),
],
check=True,
)
def is_complete(clip_dir: Path) -> bool:
return (clip_dir / "audio.wav").exists() and len(
list((clip_dir / "frames").glob("*.jpg"))
) == FRAME_COUNT
def main() -> None:
parser = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter)
parser.add_argument("--data-dir", type=Path, default=Path("data"))
parser.add_argument("--out-dir", type=Path, default=Path("media"))
parser.add_argument("--segments", type=Path, nargs="+", required=True, help="AudioSet segment CSVs")
parser.add_argument("--users", default="", help="Comma-separated user IDs (default: all)")
args = parser.parse_args()
users = {u.strip() for u in args.users.split(",") if u.strip()} or None
segments = load_segments(args.segments)
assets = sorted(session_assets(args.data_dir, users))
args.out_dir.mkdir(parents=True, exist_ok=True)
print(f"{len(assets)} clips to fetch", file=sys.stderr)
missing: list[str] = []
for number, asset_id in enumerate(assets, start=1):
clip_dir = args.out_dir / asset_id
if is_complete(clip_dir):
continue
video_id = youtube_id(asset_id)
if video_id not in segments:
missing.append(f"{asset_id}\tnot in the segment CSVs")
continue
start, end = segments[video_id]
try:
with tempfile.TemporaryDirectory() as tmp:
video = Path(tmp) / "clip.mp4"
download_segment(video_id, start, end, video)
write_media(video, clip_dir)
except subprocess.CalledProcessError as error:
missing.append(f"{asset_id}\t{error.cmd[0]} failed")
continue
print(f"[{number}/{len(assets)}] {asset_id}", file=sys.stderr)
(args.out_dir / "missing.txt").write_text("".join(f"{line}\n" for line in missing))
print(f"done; {len(missing)} clips unavailable (see {args.out_dir / 'missing.txt'})", file=sys.stderr)
if __name__ == "__main__":
main()
|