Supports syllables, 5-character sequences, and full words
Precise start-end timing for every fingerspelled unit
Includes detailed demographic and consent metadata for participant-aware modeling and analysis.
Available under terms restricting use to non-commercial academic research
An overview of the basic information and participant information included in this dataset. Provided as CSV files.
Basic information such as the video file path, category, and recording conditions.
| Field Name | Type | Description |
|---|---|---|
| file_name | str | File path of the video sample |
| classes | List[str] | Fingerspelled unit (e.g., `["a"]`, `["ka", "ma", "ku", "ra"]`) |
| category | int | Linguistic unit category: `0=syllable`, `1=sequence`, or `2=word` |
| participant_id | int | Participant identifier (e.g., `18`) |
| recording_date | int | Year and month of recording (e.g., `202403`) |
| fps | int | Frames per second (e.g., `30`) |
Anonymized attribute information of participants who cooperated in data collection.
| Field Name | Type | Description |
|---|---|---|
| participant_id | int | Participant identifier (e.g., `18`) |
| age_group | str | Age decade group (e.g., `40` for age 40-49; `-1` if not provided) |
| gender | int | Gender category: `0=female`, `1=male`, `-1` if unspecified |
| dominant_hand | int | Dominant hand: `0=right`, `1=left`, `-1` if unspecified |
| experience_years | str | Years of sign language experience: one of `1-3`, `4-6`, ..., `51+` or `-1` |
| hearing_level | int | Self-reported hearing ability: `0` (no issue) to `4` (severe), or `-1`(unknown)` |
| face_visibility | int | Face visibility consent: `1=agreed`, `0=declined` |
ub-MOJI supports temporal action detection tasks. It provides annotations in TOML files indicating the start and end positions of finger-spelling classes for each video sample. Each top-level TOML table represents a single video identified by a unique video ID. All annotations were manually performed by authors or contributors.
@misc{ubmoji2025,
title = {ub-MOJI},
author = {Tamon Kondo and Ryota Murai and Naoto Tsuta and Yousun Kang},
year = {2025},
url = {https://huggingface.co/datasets/kanglabs/ub-MOJI},
publisher = {Hugging Face}
}@inproceedings{Murai2025pointSupervisedJF,
title = {Point-Supervised Japanese Fingerspelling Localization via HR-Pro and Contrastive Learning},
author = {Ryota Murai and Naoto Tsuta and Duk Shin and Yousun Kang},
booktitle = {Proceedings of 2025 IEEE/CVF International Conference on Computer Vision Workshops (ICCVW)},
year = {2025},
}