how2sign-asl-clips

Sentence-level clips from the How2Sign ASL dataset, cut to the realigned timestamps in how2sign_realigned_train.csv. Built for the EPFL CS-503 hand2string project.

This is a work-in-progress mirror of a single shard (351 source videos, 4991 clips). More splits and shards will be added as we download them.

Schema (metadata.parquet)

column	type	notes
sentence_id	string	primary key, e.g. --7E2sU6zP4_10
sentence_name	string	full How2Sign clip name with camera tag
video_id	string	parent YouTube id
video_name	string	source mp4 basename (no extension)
start	float	timestamp in source video (s)
end	float	timestamp in source video (s)
duration	float	end - start (s)
sentence	string	English transcript
split	string	dataset split
file_name	string	clip path relative to repo root

Quick start

from huggingface_hub import snapshot_download
import pandas as pd
from pathlib import Path

local = Path(snapshot_download("martinctl/how2sign-asl-clips", repo_type="dataset"))
df    = pd.read_parquet(local / "metadata.parquet")

row = df.iloc[0]
print(row.sentence)
clip = local / row.file_name   # playable mp4

License

How2Sign is released under CC BY-NC 4.0. Cite the original authors:

Duarte, A., Palaskar, S., Ventura, L., Ghadiyaram, D., DeHaan, K., Metze, F., Torres, J., Giró-i-Nieto, X. How2Sign: A Large-scale Multimodal Dataset for Continuous American Sign Language. CVPR 2021.