Skip to content

Dataset · Linzhan

Unified text-guided 3D animation dataset for diverse skeletons

Linzhan/UniML3D

10k to 100k motion clips of humans, animals, and objects with unified skeletons and English text descriptions.

What's inside

The UniML3D dataset, created for the UniMate project (SIGGRAPH Asia 2026), combines 3D animations from three heterogeneous sources: Mixamo humanoids, Truebones ZOO animals, and Objaverse-XL rigged objects. All skeletons are converted to a unified format with cleaned joint names, body type classifications, and gaze directions, with each clip accompanied by a generated text description.

What you can do with it

  • Training text-to-3D motion generation models for any creature
  • Developing motion retargeting systems between different skeleton types
  • Training multimodal models to understand 3D character physics and kinematics
Tasks
Text → 3d
Languages
English
Size
10K–100K rows
Downloads this month
18.2K
Likes
18

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("Linzhan/UniML3D")