QFun/MANUS-HaGRID
MANUS-HaGRID: HaGRID-derived Multimodal Annotated Naturalistic Hand Understanding Dataset MANUS-HaGRID is the HaGRID/HaGRIDv2-derived subset of the Multimodal Annotated Naturalistic Hand Understanding (MANUS) dataset family. It provides multimodal annotations for naturalistic hand gesture understanding, including RGB images, hand crops, depth maps, 2D bounding boxes, estimated MANO-style hand mesh metadata, and multi-view mesh renderings where available. This repository contains… See the full description on the dataset page: https://huggingface.co/datasets/QFun/MANUS-HaGRID.
0223
1from pathlib import Path2import re3import pandas as pd4 5allowed_sources = {'HaGRID', 'HaGRIDv2'}6drive_path = re.compile(r"^[A-Za-z]:\\")7for path in Path("metadata").glob('hagrid*manifest.parquet'):8 df = pd.read_parquet(path)9 if "source_dataset" not in df.columns:10 raise ValueError(f"{path} is missing source_dataset")11 bad = set(df["source_dataset"].dropna().unique()) - allowed_sources12 if bad:13 raise ValueError(f"{path} contains unexpected sources: {bad}")14 for col in df.columns:15 values = df[col].dropna().astype(str).head(1000)16 if any(drive_path.match(v) or v.startswith("/data/") for v in values):17 raise ValueError(f"{path} column {col} appears to contain local absolute paths")18print('All MANUS-HaGRID manifests contain only HaGRID/HaGRIDv2 portable metadata.')19