MySign: A High-Fidelity Motion-Capture Dataset for 3D Sign Generation in Bahasa Isyarat Malaysia

MySign is a high-fidelity motion-capture dataset for 3D sign generation in Bahasa Isyarat Malaysia (BIM) — Malaysian Sign Language. It contains skeletal animations of 1,194 unique BIM vocabulary items performed by 5 deaf signers (Signer001–Signer005), totalling 5,343 motion-capture clips stored in Autodesk FBX format.

The file name of each .fbx encodes the English gloss for the corresponding BIM sign.

Structure

MySign/
├── Signer001/{GLOSS}.fbx     # ~1068 signs per signer
├── Signer002/{GLOSS}.fbx
├── Signer003/{GLOSS}.fbx
├── Signer004/{GLOSS}.fbx
├── Signer005/{GLOSS}.fbx
├── metadata.csv              # gloss / signer / take index
└── croissant.json            # ML Commons Croissant metadata

metadata.csv columns

column	description
file_name	Relative path of the FBX file (e.g. Signer001/Above.fbx)
gloss	English gloss label for the BIM sign; synonyms separated by /
signer_id	One of Signer001 .. Signer005
take	Take number; defaults to 1 unless multiple recordings exist

Loading

The FBX files are 3D motion-capture recordings and require a tool such as Blender, Autodesk FBX SDK, or pyfbx to parse. The metadata.csv provides the mapping from each file to its gloss label and signer.

import pandas as pd
from huggingface_hub import hf_hub_download

# Load the index
df = pd.read_csv(
    hf_hub_download("mysigner/MySign", "metadata.csv", repo_type="dataset")
)

# Download a single FBX (e.g. the BIM sign for "ABOVE" by Signer001)
row = df.query("gloss == 'ABOVE' and signer_id == 'Signer001'").iloc[0]
fbx_path = hf_hub_download("mysigner/MySign", row["file_name"], repo_type="dataset")
print(f"Downloaded to: {fbx_path}")

Intended use

MySign is intended to support research on:

• 3D sign language generation and synthesis for BIM
• Isolated Sign Language recognition (ISLR)
• Sign language animation, avatars, and accessibility tools for the Malaysian Deaf community

License

CC BY-NC-SA 4.0 — non-commercial use, with attribution and same-license sharing.

Citation

@misc{mysign2026,
  title  = {MySign: A High-Fidelity Motion-Capture Dataset for 3D Sign Generation in Bahasa Isyarat Malaysia},
  author = {mysigner},
  year   = {2026},
  url    = {https://huggingface.co/datasets/mysigner/MySign}
}