Mingdian Liu

Title: Mingdian Liu and Text-to-Motion Animation Innovation.

Introduction

Mingdian Liu is an inventor based in Ames, Iowa, and works at Snap Inc. His patent work focuses on generating expressive text-to-motion animations from partially annotated datasets.

Latest Patents

Liu is associated with one patent, titled “Generating Text-to-Motion Animations from Partially Annotated Datasets.” The invention describes T2M-X, a two-stage approach for learning and generating expressive animations from datasets with incomplete annotations.

In the first stage, T2M-X combines partially annotated datasets into a unified motion dataset and trains separate vector-quantized variational autoencoders for body, hand, and face motion. These models generate high-quality motion outputs. In the second stage, a multi-indexing generative pre-trained transformer learns from those outputs, using motion-consistency loss and sequence-length consistency to generate coordinated, expressive animations.

Career Highlights

Liu’s disclosed invention addresses the challenge of learning from motion datasets that are only partially annotated. Its two-stage design brings body, hand, and facial motion together to support coordinated text-to-motion animation generation.

Collaborations

Liu’s listed company affiliation is Snap Inc. The provided information does not specify additional collaborators.

Conclusion

Mingdian Liu’s patent work presents a two-stage method for turning text into coordinated, expressive motion animations using partially annotated datasets, and he is associated with 1 patent.