VidTouch / scripts /prepare_release.py
AQUILA-espresso's picture
Release canonical VidTouch dataset v1.0.0
299146f verified
Raw
History Blame Contribute Delete
9.62 kB
from __future__ import annotations
import argparse
import csv
import hashlib
import json
import os
import re
import shutil
from collections import Counter
from pathlib import Path
from typing import Any
IMAGE_EXTENSIONS = {".jpg", ".jpeg", ".png", ".bmp", ".webp"}
VIDEO_EXTENSIONS = {".mp4", ".mov", ".avi", ".mkv"}
ATTRIBUTES = ("weave", "material", "usage", "features")
def parse_args() -> argparse.Namespace:
parser = argparse.ArgumentParser(description="Build the canonical VidTouch Hub release.")
parser.add_argument("--source", type=Path, required=True)
parser.add_argument("--output", type=Path, required=True)
return parser.parse_args()
def parse_labels(path: Path) -> dict[str, dict[str, Any]]:
labels: dict[str, dict[str, Any]] = {}
with path.open("r", encoding="utf-8") as handle:
for line_number, raw in enumerate(handle, 1):
line = raw.strip()
if not line or line.startswith("#"):
continue
parts = line.split()
if len(parts) < 4:
raise ValueError(f"Invalid label line {line_number}: {raw!r}")
fabric_id, weave, material, usage, *features = parts
if fabric_id in labels:
raise ValueError(f"Duplicate Fabric ID in labels: {fabric_id}")
labels[fabric_id] = {
"fabric_id": fabric_id,
"weave": weave,
"material": material,
"usage": usage,
"features": features,
}
return labels
def parse_fabric_id(path: Path) -> str:
match = re.match(r"^([A-Za-z0-9]+)", path.stem)
if not match:
raise ValueError(f"Cannot parse Fabric ID from {path.name}")
return match.group(1)
def sha256(path: Path) -> str:
digest = hashlib.sha256()
with path.open("rb") as handle:
for chunk in iter(lambda: handle.read(8 * 1024 * 1024), b""):
digest.update(chunk)
return digest.hexdigest()
def link_or_copy(source: Path, destination: Path) -> None:
destination.parent.mkdir(parents=True, exist_ok=True)
if destination.exists():
destination.unlink()
try:
os.link(source, destination)
except OSError:
shutil.copy2(source, destination)
def scan_media(
folder: Path,
extensions: set[str],
labels: dict[str, dict[str, Any]],
) -> tuple[list[Path], list[str]]:
retained: list[Path] = []
excluded: list[str] = []
for path in sorted(folder.iterdir(), key=lambda item: item.name):
if not path.is_file() or path.suffix.lower() not in extensions:
continue
if parse_fabric_id(path) in labels:
retained.append(path)
else:
excluded.append(path.name)
return retained, excluded
def write_csv(path: Path, fieldnames: list[str], rows: list[dict[str, Any]]) -> None:
path.parent.mkdir(parents=True, exist_ok=True)
with path.open("w", encoding="utf-8", newline="") as handle:
writer = csv.DictWriter(handle, fieldnames=fieldnames)
writer.writeheader()
writer.writerows(rows)
def main() -> None:
args = parse_args()
source = args.source.resolve()
output = args.output.resolve()
labels_path = source / "label.txt"
split_dir = source / "experiments" / "vidtouch_method" / "splits"
labels = parse_labels(labels_path)
split = json.loads((split_dir / "fabric_common_v2.json").read_text(encoding="utf-8"))
partition_by_id: dict[str, str] = {}
for partition, key in (("train", "train_ids"), ("validation", "val_ids"), ("test", "test_ids")):
for fabric_id in split[key]:
if fabric_id in partition_by_id:
raise ValueError(f"Fabric ID appears in multiple partitions: {fabric_id}")
partition_by_id[fabric_id] = partition
if set(partition_by_id) != set(labels):
raise ValueError("Frozen split does not cover exactly the canonical Fabric IDs.")
images, excluded_images = scan_media(source / "RGBs", IMAGE_EXTENSIONS, labels)
videos, excluded_videos = scan_media(source / "TACs", VIDEO_EXTENSIONS, labels)
image_counts = Counter(parse_fabric_id(path) for path in images)
video_counts = Counter(parse_fabric_id(path) for path in videos)
missing_images = sorted(set(labels) - set(image_counts))
missing_videos = sorted(set(labels) - set(video_counts))
if missing_images or missing_videos:
raise ValueError(
f"Missing media: RGB={missing_images}, tactile={missing_videos}"
)
expected = {
"fabrics": 144,
"rgb_images": 435,
"tactile_videos": 432,
"partitions": {"train": 100, "validation": 22, "test": 22},
"label_cardinality": {"weave": 39, "material": 60, "usage": 46, "features": 100},
}
label_cardinality = {
"weave": len({row["weave"] for row in labels.values()}),
"material": len({row["material"] for row in labels.values()}),
"usage": len({row["usage"] for row in labels.values()}),
"features": len({feature for row in labels.values() for feature in row["features"]}),
}
actual = {
"fabrics": len(labels),
"rgb_images": len(images),
"tactile_videos": len(videos),
"partitions": dict(Counter(partition_by_id.values())),
"label_cardinality": label_cardinality,
}
if actual != expected:
raise ValueError(f"Release statistics differ from the frozen specification: {actual}")
for relative in (
"RGBs",
"TACs",
"metadata",
"splits",
):
(output / relative).mkdir(parents=True, exist_ok=True)
link_or_copy(labels_path, output / "label.txt")
for split_name in (
"fabric_common_v2.json",
"fabric_common_v2_lowshot25.json",
"fabric_common_v2_lowshot50.json",
):
link_or_copy(split_dir / split_name, output / "splits" / split_name)
fabric_rows: list[dict[str, Any]] = []
for fabric_id in sorted(labels):
label = labels[fabric_id]
fabric_rows.append(
{
"fabric_id": fabric_id,
"split": partition_by_id[fabric_id],
"weave": label["weave"],
"material": label["material"],
"usage": label["usage"],
"features": json.dumps(label["features"], ensure_ascii=True),
"rgb_count": image_counts[fabric_id],
"tactile_count": video_counts[fabric_id],
}
)
write_csv(
output / "metadata" / "fabrics.csv",
["fabric_id", "split", "weave", "material", "usage", "features", "rgb_count", "tactile_count"],
fabric_rows,
)
observation_rows: list[dict[str, Any]] = []
for modality, paths, destination_name in (
("rgb", images, "RGBs"),
("tactile", videos, "TACs"),
):
modality_rows: list[dict[str, Any]] = []
for path in paths:
fabric_id = parse_fabric_id(path)
label = labels[fabric_id]
link_or_copy(path, output / destination_name / path.name)
row = {
"file_name": path.name,
"fabric_id": fabric_id,
"split": partition_by_id[fabric_id],
"weave": label["weave"],
"material": label["material"],
"usage": label["usage"],
"features": json.dumps(label["features"], ensure_ascii=True),
}
modality_rows.append(row)
observation_rows.append(
{
"path": f"{destination_name}/{path.name}",
"modality": modality,
**{key: value for key, value in row.items() if key != "file_name"},
}
)
write_csv(
output / destination_name / "metadata.csv",
["file_name", "fabric_id", "split", "weave", "material", "usage", "features"],
modality_rows,
)
write_csv(
output / "metadata" / "observations.csv",
["path", "modality", "fabric_id", "split", "weave", "material", "usage", "features"],
observation_rows,
)
release_manifest = {
"release_name": "VidTouch canonical release",
"release_version": "1.0.0",
"statistics": actual,
"rgb_per_fabric_distribution": dict(sorted(Counter(image_counts.values()).items())),
"tactile_per_fabric_distribution": dict(sorted(Counter(video_counts.values()).items())),
"excluded_unannotated_source_media": {
"rgb": excluded_images,
"tactile": excluded_videos,
},
"canonical_label_sha256": sha256(labels_path),
"frozen_split_sha256": sha256(split_dir / "fabric_common_v2.json"),
"assignment_sha256": split["metadata"]["assignment_sha256"],
"data_manifest_sha256": split["metadata"]["data_manifest_sha256"],
}
(output / "release_manifest.json").write_text(
json.dumps(release_manifest, indent=2, ensure_ascii=True) + "\n",
encoding="utf-8",
)
checksum_paths = sorted(
path
for path in output.rglob("*")
if path.is_file() and path.name != "checksums.sha256"
)
with (output / "checksums.sha256").open("w", encoding="utf-8", newline="\n") as handle:
for path in checksum_paths:
relative = path.relative_to(output).as_posix()
handle.write(f"{sha256(path)} {relative}\n")
print(json.dumps(release_manifest, indent=2))
if __name__ == "__main__":
main()