The foundational paper for this dataset, which describes its collection, annotation, and metadata, is:
This paper introduces the dataset containing over 80 hours of instructional videos, including the BFH (Body-Face-Hands) keypoint annotations mentioned in your file name. You can access more information via the How2Sign Project Page . Dataset Context BFH03_03.mp4
If your query refers to the specifically, they produce significant research in applied machine intelligence and video analysis: The foundational paper for this dataset, which describes
The video file is a specific clip from the How2Sign dataset , a large-scale multimodal collection of American Sign Language (ASL) videos designed for computer vision research. Primary Research Paper Primary Research Paper Authors: Amanda Duarte, et al
Authors: Amanda Duarte, et al. Venue: CVPR 2021
are specialized subsets that include these detailed pose estimations for use in tasks like sign language translation and human pose regression.
The naming convention "BFH" typically refers to keypoints, which are high-resolution 2D or 3D skeletal data extracted from RGB videos. In the How2Sign dataset: