Deep learning techniques have been widely developed to process 3D data directly from point clouds. It presents a comparative analysis of PointNet++, a Transformer-based reconstruction model, and a proposed hybrid framework that integrates both approaches for 3D reconstruction from imperfect point cloud data. The evaluation is conducted based on geometric accuracy, structural detail preservation, and surface quality. These findings highlight the complementary nature of local and global representations and support the selection of hybrid strategies for accurate and scalable medical 3D reconstruction.
مشاريع مقترحة