<pstyle="text-align: center;">This diagram illustrates the progressive unmasking of joints in the 3D skeleton and the gradual expansion of model layers within the self-supervised learning framework using Low-Rank Adaptation (LoRA). The approach aims to enhance feature learning while maintaining a manageable model size, allowing for more complex motion capture over time.</p>
<pstyle="text-align: center;">This figure presents the architecture of our method, showcasing the model's structure and the integration of LoRA layers. The diagram highlights how the model evolves, adding layers to capture increasingly complex motions while leveraging the benefits of self-supervised learning.</p>
<pstyle="text-align: center;">This figure displays the performance of our model using LoRA, comparing its efficiency and effectiveness in capturing 3D skeleton representations. The graph emphasizes the improvements in feature learning while maintaining a small model size, demonstrating the strength of LoRA in capturing robust skeleton representations.</p>
