SignAvatar: Sign Language 3D Motion Reconstruction and Generation

TL;DR

SignAvatar is a novel framework that reconstructs and generates expressive 3D sign language motions at the word level, leveraging a transformer-based autoencoder and curriculum learning, supported by a new 3D sign language dataset.

Contribution

We introduce SignAvatar, a transformer-based autoencoder for 3D sign language reconstruction and generation, and provide the ASL3DWord dataset for isolated sign words.

Findings

SignAvatar achieves superior reconstruction accuracy.

The model effectively generates realistic sign language motions.

The dataset enables better training and evaluation of sign language models.

Abstract

Achieving expressive 3D motion reconstruction and automatic generation for isolated sign words can be challenging, due to the lack of real-world 3D sign-word data, the complex nuances of signing motions, and the cross-modal understanding of sign language semantics. To address these challenges, we introduce SignAvatar, a framework capable of both word-level sign language reconstruction and generation. SignAvatar employs a transformer-based conditional variational autoencoder architecture, effectively establishing relationships across different semantic modalities. Additionally, this approach incorporates a curriculum learning strategy to enhance the model's robustness and generalization, resulting in more realistic motions. Furthermore, we contribute the ASL3DWord dataset, composed of 3D joint rotation data for the body, hands, and face, for unique sign words. We demonstrate the effectiveness of SignAvatar through extensive experiments, showcasing its superior reconstruction and automatic generation capabilities. The code and dataset are available on the project page.