Spaces:
Sleeping
Sleeping
File size: 18,572 Bytes
5412d82 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 | """
ARKitScenes subset loader for the depth-aware scene description evaluation.
Downloads a small, reproducible set of indoor RGB frames from the ARKitScenes
Validation split [Baruch et al., NeurIPS 2021]. Frames are the evaluation
dataset for all three ablation metrics (STD, SFS, Preamble BERTScore).
Dataset:
Baruch, G., Chen, Z., Dehghan, A., et al. (2021). ARKitScenes β A Diverse
Real-World Dataset for 3D Indoor Scene Understanding.
NeurIPS 2021 Datasets and Benchmarks Track.
https://github.com/apple/ARKitScenes
Prerequisites
-------------
Clone the ARKitScenes repository and set the ``ARKITSCENES_REPO`` environment
variable to its absolute path, OR pass ``--repo-dir`` on the CLI::
git clone https://github.com/apple/ARKitScenes /path/to/ARKitScenes
set ARKITSCENES_REPO=C:\\path\\to\\ARKitScenes (Windows CMD)
$env:ARKITSCENES_REPO = "C:\\path\\to\\ARKitScenes" (Windows PowerShell)
export ARKITSCENES_REPO=/path/to/ARKitScenes (Linux/macOS)
The repo is cloned only to read the splits CSV. Frame downloads go directly
to the Apple CDN via Python builtins (no curl or unzip shell commands).
Downloaded frames land at::
<tmp_dir>/raw/Validation/<video_id>/lowres_wide/*.png
Usage (library)::
from src.data.arkitscenes_loader import fetch_arkitscenes_subset
paths = fetch_arkitscenes_subset(n_videos=20, frames_per_video=10)
CLI::
python -m src.data.arkitscenes_loader \\
--n-videos 20 --frames-per-video 10 --output data/test_images/ \\
--repo-dir C:\\path\\to\\ARKitScenes
"""
from __future__ import annotations
import argparse
import csv
import os
import random
import tempfile
import urllib.request
import zipfile
from pathlib import Path
from typing import Optional
from ..config import TEST_IMAGES_DIR
# ---------------------------------------------------------------------------
# Constants
# ---------------------------------------------------------------------------
# Path of the splits manifest *inside* the ARKitScenes repo clone.
_SPLITS_REL = Path("raw") / "raw_train_val_splits.csv"
# Column names inside the CSV (ARKitScenes convention).
_COL_VIDEO_ID = "video_id"
_COL_FOLD = "fold"
_VALIDATION_FOLD = "Validation"
# ARKitScenes CDN base URL (public, no auth required).
_CDN_BASE = "https://docs-assets.developer.apple.com/ml-research/datasets/arkitscenes/v1"
# lowres_wide = 256Γ192 wide-angle RGB frames, sufficient for VLM evaluation.
_RGB_ASSET = "lowres_wide"
# Image extensions to accept when scanning downloaded frames.
_IMAGE_EXTS = {".png", ".jpg", ".jpeg"}
# Environment variable users set to point at their ARKitScenes clone.
_REPO_ENV_VAR = "ARKITSCENES_REPO"
def _default_tmp_dir() -> Path:
"""Return a system temp directory guaranteed to have no spaces in its path.
``download_data.py`` invokes curl/unzip via shell strings without quoting
paths, which breaks on Windows when the project directory contains spaces
(e.g. ``C:\\Users\\risha\\New folder\\...``). We avoid that entirely by
downloading and extracting with Python builtins, but we still want the
scratch directory to be clean and outside the project tree.
"""
return Path(tempfile.gettempdir()) / "arkit_tmp"
# ---------------------------------------------------------------------------
# Internal helpers
# ---------------------------------------------------------------------------
def _read_splits_csv(repo_dir: Path) -> list[dict[str, str]]:
"""Read the train/val splits manifest from the local ARKitScenes clone.
Args:
repo_dir: Root of the cloned ARKitScenes repository.
Returns:
List of row dicts, one per video in the manifest.
Raises:
FileNotFoundError: If the CSV is not found at the expected path inside
the repo clone.
"""
csv_path = repo_dir / _SPLITS_REL
if not csv_path.exists():
raise FileNotFoundError(
f"Splits CSV not found at {csv_path}.\n"
f"Make sure --repo-dir (or ${_REPO_ENV_VAR}) points to the root "
f"of a complete ARKitScenes clone."
)
print(f" Reading splits CSV: {csv_path}")
with open(csv_path, newline="", encoding="utf-8") as fh:
reader = csv.DictReader(fh)
rows = list(reader)
return rows
def _validation_video_ids(rows: list[dict[str, str]]) -> list[str]:
"""Extract video IDs for the Validation fold.
Args:
rows: All rows from the splits CSV.
Returns:
Sorted list of video ID strings marked as Validation.
Raises:
ValueError: If the expected column names are absent.
"""
if not rows:
raise ValueError("Splits CSV is empty.")
first = rows[0]
if _COL_VIDEO_ID not in first or _COL_FOLD not in first:
raise ValueError(
f"Splits CSV columns do not match expected schema. "
f"Got: {list(first.keys())}. "
f"Expected '{_COL_VIDEO_ID}' and '{_COL_FOLD}'."
)
ids = [r[_COL_VIDEO_ID] for r in rows if r[_COL_FOLD] == _VALIDATION_FOLD]
if not ids:
raise ValueError(
f"No rows with fold='{_VALIDATION_FOLD}' found in splits CSV."
)
return sorted(ids)
def _resolve_repo_dir(repo_dir: Optional[Path]) -> Path:
"""Return the ARKitScenes repo directory, checking env var as fallback.
The repo is only needed to read the splits CSV. It is NOT used for
downloading frames (we fetch directly from the Apple CDN with Python
builtins to avoid Windows curl/unzip path-with-spaces issues).
Args:
repo_dir: Explicit path supplied by the caller (may be None).
Returns:
Resolved Path to the ARKitScenes clone root.
Raises:
RuntimeError: If neither argument nor env var points to a valid dir.
"""
if repo_dir is not None:
resolved = Path(repo_dir).expanduser().resolve()
if resolved.is_dir():
return resolved
raise RuntimeError(f"--repo-dir does not exist: {resolved}")
env_val = os.environ.get(_REPO_ENV_VAR)
if env_val:
resolved = Path(env_val).expanduser().resolve()
if resolved.is_dir():
return resolved
raise RuntimeError(
f"${_REPO_ENV_VAR} is set to '{env_val}' but that directory "
f"does not exist."
)
raise RuntimeError(
f"ARKitScenes repo directory not found.\n"
f"Either:\n"
f" 1. Pass --repo-dir on the command line:\n"
f" --repo-dir \"C:\\path\\to\\ARKitScenes\"\n"
f" 2. Set the {_REPO_ENV_VAR} environment variable:\n"
f" Windows CMD: set {_REPO_ENV_VAR}=C:\\path\\to\\ARKitScenes\n"
f" Windows PowerShell: $env:{_REPO_ENV_VAR} = \"C:\\path\\to\\ARKitScenes\"\n"
f" Linux/macOS: export {_REPO_ENV_VAR}=/path/to/ARKitScenes\n"
f"\n"
f"Clone with: git clone https://github.com/apple/ARKitScenes"
)
def _download_video(video_id: str, tmp_dir: Path) -> Optional[Path]:
"""Download and extract the lowres_wide RGB frames for one video_id.
Downloads the zip directly from the Apple ARKitScenes CDN using
``urllib.request`` and extracts it with ``zipfile``. This is fully
cross-platform and handles paths with spaces correctly β unlike the
``download_data.py`` approach which uses unquoted curl/unzip shell
commands that break on Windows paths containing spaces.
The zip is streamed to ``<tmp_dir>/<video_id>.zip``, then extracted to
``<tmp_dir>/<video_id>/``.
Args:
video_id: ARKitScenes video identifier string (e.g. ``"42444596"``).
tmp_dir: Scratch directory to write the zip and extracted frames.
Must be a path **without spaces** (use the system temp dir).
Returns:
Path to the extracted frame directory, or None on failure.
"""
url = f"{_CDN_BASE}/raw/{_VALIDATION_FOLD}/{video_id}/{_RGB_ASSET}.zip"
zip_path = tmp_dir / f"{video_id}.zip"
extract_dir = tmp_dir / video_id
# ββ Download ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
if not zip_path.exists():
print(f" Downloading {video_id} from CDN β¦", flush=True)
try:
def _progress(block_num: int, block_size: int, total: int) -> None:
if total > 0:
pct = min(100, block_num * block_size * 100 // total)
print(f"\r {pct:3d}% ({block_num * block_size / 1e6:.1f} MB)", end="", flush=True)
urllib.request.urlretrieve(url, zip_path, reporthook=_progress)
print() # newline after progress
except Exception as exc:
print(f"\n [WARN] Download failed for {video_id}: {exc}")
if zip_path.exists():
zip_path.unlink()
return None
else:
print(f" {video_id}.zip already downloaded, skipping.")
# ββ Extract βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
if not extract_dir.exists():
print(f" Extracting {video_id}.zip β¦", flush=True)
try:
with zipfile.ZipFile(zip_path, "r") as zf:
zf.extractall(extract_dir)
except zipfile.BadZipFile as exc:
print(f" [WARN] Bad zip for {video_id}: {exc}")
zip_path.unlink(missing_ok=True)
return None
else:
print(f" {video_id}/ already extracted, skipping.")
return extract_dir
def _collect_frames(video_dir: Path) -> list[Path]:
"""Recursively collect all image files under a video directory, sorted.
Args:
video_dir: Root directory to search.
Returns:
Sorted list of image Paths.
"""
return sorted(
p for p in video_dir.rglob("*")
if p.suffix.lower() in _IMAGE_EXTS and p.is_file()
)
def _evenly_spaced(items: list, n: int) -> list:
"""Return n evenly-spaced elements from items (not random).
Uses integer linspace so the selection is deterministic and reproducible
regardless of list length. If ``len(items) <= n``, returns all items.
Args:
items: Source list.
n: Desired number of elements.
Returns:
Selected sub-list, preserving original order.
"""
if len(items) <= n:
return list(items)
step = len(items) / n
indices = [int(i * step) for i in range(n)]
return [items[idx] for idx in indices]
# ---------------------------------------------------------------------------
# Public API
# ---------------------------------------------------------------------------
def fetch_arkitscenes_subset(
n_videos: int = 20,
frames_per_video: int = 10,
seed: int = 42,
output_dir: Path = TEST_IMAGES_DIR,
repo_dir: Optional[Path] = None,
tmp_dir: Optional[Path] = None,
skip_existing: bool = True,
) -> list[Path]:
"""Download a small, reproducible ARKitScenes Validation subset.
Reads the official Validation split manifest from the local ARKitScenes
clone, samples ``n_videos`` video IDs with a fixed seed, downloads the
``lowres_wide`` RGB asset for each video directly from the Apple CDN
using Python builtins (no curl, no unzip shell commands), then copies
``frames_per_video`` evenly-spaced frames to ``output_dir`` with
canonical names (``arkit_<video_id>_<frame_idx:04d>.jpg``).
Args:
n_videos: Number of video sequences to sample.
frames_per_video: Frames to keep per sequence (evenly spaced,
deterministic β not random).
seed: Random seed for video sampling. Fixed at 42 in
all published results for reproducibility.
output_dir: Destination directory for the sampled JPEG frames.
repo_dir: Path to a local ARKitScenes clone (needed only to
read the splits CSV). If None, the
``ARKITSCENES_REPO`` environment variable is used.
tmp_dir: Scratch directory for zip downloads and extraction.
Defaults to ``<system_temp>/arkit_tmp`` to avoid
path-with-spaces issues on Windows. Safe to delete
after this function returns.
skip_existing: If True, skip frames whose destination path already
exists (allows resuming interrupted runs).
Returns:
Sorted list of output JPEG Paths (up to n_videos Γ frames_per_video).
Raises:
RuntimeError: If the ARKitScenes repo directory cannot be resolved.
FileNotFoundError: If the splits CSV is missing from the repo clone.
"""
output_dir = Path(output_dir)
# Default tmp_dir to system temp to guarantee a spaceless path on Windows.
resolved_tmp = Path(tmp_dir) if tmp_dir is not None else _default_tmp_dir()
output_dir.mkdir(parents=True, exist_ok=True)
resolved_tmp.mkdir(parents=True, exist_ok=True)
resolved_repo = _resolve_repo_dir(repo_dir)
print(f"ARKitScenes repo : {resolved_repo}")
print(f"Output directory : {output_dir}")
print(f"Temp directory : {resolved_tmp}")
print(f"Seed={seed} n_videos={n_videos} frames_per_video={frames_per_video}")
print()
# ββ 1. Read validation split from local repo ββββββββββββββββββββββββββββββ
rows = _read_splits_csv(resolved_repo)
val_ids = _validation_video_ids(rows)
print(f" Validation videos in manifest: {len(val_ids)}")
# ββ 2. Sample n_videos with fixed seed ββββββββββββββββββββββββββββββββββββ
rng = random.Random(seed)
sampled_ids = sorted(rng.sample(val_ids, min(n_videos, len(val_ids))))
print(f" Sampled {len(sampled_ids)} videos (seed={seed}):")
for vid in sampled_ids:
print(f" {vid}")
print()
# ββ 3. Download + sample each video ββββββββββββββββββββββββββββββββββββββ
output_paths: list[Path] = []
skipped_videos = 0
for idx, video_id in enumerate(sampled_ids, start=1):
print(f"[{idx}/{len(sampled_ids)}] video_id={video_id}", flush=True)
frames_dir = _download_video(video_id, resolved_tmp)
if frames_dir is None:
print(f" Skipped (download failed).")
skipped_videos += 1
continue
all_frames = _collect_frames(frames_dir)
if not all_frames:
print(f" Skipped (no image files found under {frames_dir}).")
skipped_videos += 1
continue
print(f" {len(all_frames)} frames found β keeping {frames_per_video}")
selected = _evenly_spaced(all_frames, frames_per_video)
for frame_path in selected:
frame_idx = all_frames.index(frame_path)
dest_name = f"arkit_{video_id}_{frame_idx:04d}.jpg"
dest = output_dir / dest_name
if skip_existing and dest.exists():
print(f" {dest_name} (exists, skipped)")
output_paths.append(dest)
continue
try:
from PIL import Image as _PILImage # noqa: PLC0415
img = _PILImage.open(frame_path).convert("RGB")
img.save(dest, format="JPEG", quality=95)
print(f" {dest_name}")
output_paths.append(dest)
except Exception as exc:
print(f" [WARN] Could not save {frame_path.name}: {exc}")
# ββ 4. Summary ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
print()
print(f"Done. {len(output_paths)} frames written to {output_dir}")
if skipped_videos:
print(f" ({skipped_videos} videos skipped)")
return sorted(output_paths)
# ---------------------------------------------------------------------------
# CLI
# ---------------------------------------------------------------------------
def _parse_args(argv: list[str] | None = None) -> argparse.Namespace:
p = argparse.ArgumentParser(
description=(
"Download a reproducible ARKitScenes Validation subset for "
"depth-aware scene description evaluation."
)
)
p.add_argument(
"--n-videos", type=int, default=20,
help="Number of video sequences to sample (default: 20).",
)
p.add_argument(
"--frames-per-video", type=int, default=10,
help="Evenly-spaced frames to keep per video (default: 10).",
)
p.add_argument(
"--seed", type=int, default=42,
help="Random seed for video sampling (default: 42). "
"Keep at 42 to match published results.",
)
p.add_argument(
"--output", type=Path, default=TEST_IMAGES_DIR,
help=f"Output directory for JPEG frames (default: {TEST_IMAGES_DIR}).",
)
p.add_argument(
"--repo-dir", type=Path, default=None,
help=(
"Path to the cloned ARKitScenes repo. "
f"Overrides the ${_REPO_ENV_VAR} environment variable."
),
)
p.add_argument(
"--tmp-dir", type=Path, default=None,
help=(
"Scratch directory for zip downloads and extraction. "
"Defaults to <system_temp>/arkit_tmp to avoid Windows path-with-spaces issues."
),
)
p.add_argument(
"--no-skip-existing", action="store_true",
help="Overwrite frames that already exist in --output.",
)
return p.parse_args(argv)
def main(argv: list[str] | None = None) -> None:
"""CLI entry point."""
args = _parse_args(argv)
paths = fetch_arkitscenes_subset(
n_videos=args.n_videos,
frames_per_video=args.frames_per_video,
seed=args.seed,
output_dir=args.output,
repo_dir=args.repo_dir,
tmp_dir=args.tmp_dir,
skip_existing=not args.no_skip_existing,
)
print(f"\n{len(paths)} evaluation frames ready.")
if __name__ == "__main__":
main()
|