HMVLM: Human Motion-Vision-Lanuage Model via MoE LoRA

TL;DR

This paper introduces HMVLM, a unified human motion-vision-language model that uses MoE LoRA to enhance multimodal understanding, address catastrophic forgetting, and improve pose representation for diverse downstream tasks.

Contribution

The paper proposes a novel MoE LoRA-based framework with a zero expert and body-part-specific tokenization to improve multimodal learning and mitigate forgetting in human motion-language models.

Findings

Effectively alleviates catastrophic forgetting during instruction-tuning.

Achieves state-of-the-art performance across multiple human motion tasks.

Enhances pose representation with body-part-specific tokenization.

Abstract

The expansion of instruction-tuning data has enabled foundation language models to exhibit improved instruction adherence and superior performance across diverse downstream tasks. Semantically-rich 3D human motion is being progressively integrated with these foundation models to enhance multimodal understanding and cross-modal generation capabilities. However, the modality gap between human motion and text raises unresolved concerns about catastrophic forgetting during this integration. In addition, developing autoregressive-compatible pose representations that preserve generalizability across heterogeneous downstream tasks remains a critical technical barrier. To address these issues, we propose the Human Motion-Vision-Language Model (HMVLM), a unified framework based on the Mixture of Expert Low-Rank Adaption(MoE LoRA) strategy. The framework leverages the gating network to dynamically allocate LoRA expert weights based on the input prompt, enabling synchronized fine-tuning of multiple tasks. To mitigate catastrophic forgetting during instruction-tuning, we introduce a novel zero expert that preserves the pre-trained parameters for general linguistic tasks. For pose representation, we implement body-part-specific tokenization by partitioning the human body into different joint groups, enhancing the spatial resolution of the representation. Experiments show that our method effectively alleviates knowledge forgetting during instruction-tuning and achieves remarkable performance across diverse human motion downstream tasks.