Multi-Task Reconstruction Strategies for Unsupervised 3D Domain Adaptation

This thesis addresses the challenge of domain shift in 3D point cloud data through innovative Unsupervised Domain Adaptation (UDA) methods. Focusing on enhancing existing techniques for point cloud classification to minimize distribution shifts, the research integrates insights from UDA approaches and deep learning for 3D point cloud representations. The proposed methods are rigorously evaluated on benchmark datasets, demonstrating competitive performance with existing UDA methods. The findings contribute to advancing UDA methodologies for 3D point clouds, providing valuable insights for future research and applications in real-world scenarios.