Datasets:
File size: 3,471 Bytes
71ee09b | 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 | """
Build Parquet files for HuggingFace dataset viewer.
Reads Lightning Pose CSVs and corresponding PNG frames, then writes
Parquet files with embedded images to data/ for the HF viewer.
Usage (from repo root):
python scripts/build_parquet.py
"""
import csv
import io
from pathlib import Path
import datasets
from datasets import Dataset, Features, Image, Value, Sequence
REPO_ROOT = Path(__file__).parent.parent
DATA_OUT = REPO_ROOT / "data"
VIEWS = ["Cam-A", "Cam-B", "Cam-C", "Cam-D", "Cam-E", "Cam-F"]
KEYPOINTS = [
"L1A", "L1B", "L1C", "L1D", "L1E",
"L2A", "L2B", "L2C", "L2D", "L2E",
"L3A", "L3B", "L3C", "L3D", "L3E",
"R1A", "R1B", "R1C", "R1D", "R1E",
"R2A", "R2B", "R2C", "R2D", "R2E",
"R3A", "R3B", "R3C", "R3D", "R3E",
]
FEATURES = Features(
{
"image": Image(),
"session": Value("string"),
"view": Value("string"),
"split": Value("string"),
"frame": Value("string"),
**{f"{kp}_x": Value("float32") for kp in KEYPOINTS},
**{f"{kp}_y": Value("float32") for kp in KEYPOINTS},
}
)
def parse_csv(csv_path: Path, view: str, split: str) -> list[dict]:
rows = []
with open(csv_path) as f:
reader = csv.reader(f)
# Skip 3-row header: scorer, bodyparts, coords
next(reader)
next(reader)
next(reader)
for row in reader:
img_rel_path = row[0]
img_path = REPO_ROOT / img_rel_path
if not img_path.exists():
print(f" WARNING: missing {img_path}, skipping")
continue
coords = row[1:] # 60 values: x0,y0,x1,y1,...
record: dict = {
"image": {"path": None, "bytes": img_path.read_bytes()},
"session": "_".join(Path(img_rel_path).parent.name.split("_")[:-1]),
"view": view,
"split": split,
"frame": Path(img_rel_path).name,
}
for i, kp in enumerate(KEYPOINTS):
x_str = coords[i * 2]
y_str = coords[i * 2 + 1]
record[f"{kp}_x"] = float(x_str) if x_str else float("nan")
record[f"{kp}_y"] = float(y_str) if y_str else float("nan")
rows.append(record)
return rows
def build_split(csv_suffix: str, split_name: str) -> list[dict]:
all_rows = []
for view in VIEWS:
csv_path = REPO_ROOT / f"CollectedData_{view}{csv_suffix}.csv"
if not csv_path.exists():
print(f"Skipping missing {csv_path}")
continue
print(f" Reading {csv_path.name} ...")
rows = parse_csv(csv_path, view, split_name)
print(f" {len(rows)} rows")
all_rows.extend(rows)
return all_rows
def main():
DATA_OUT.mkdir(exist_ok=True)
print("Building InD split ...")
ind_rows = build_split("", "ind")
ind_ds = Dataset.from_list(ind_rows, features=FEATURES)
out = DATA_OUT / "ind-train-00000-of-00001.parquet"
ind_ds.to_parquet(str(out))
print(f"Wrote {out} ({out.stat().st_size / 1e6:.1f} MB, {len(ind_rows)} rows)")
print("Building OOD split ...")
ood_rows = build_split("_new", "ood")
ood_ds = Dataset.from_list(ood_rows, features=FEATURES)
out = DATA_OUT / "ood-test-00000-of-00001.parquet"
ood_ds.to_parquet(str(out))
print(f"Wrote {out} ({out.stat().st_size / 1e6:.1f} MB, {len(ood_rows)} rows)")
if __name__ == "__main__":
main()
|