DcMatch: Unsupervised Multi-Shape Matching with Dual-Level Consistency

The introduction of 'DcMatch' marks a significant advancement in the field of computer vision and graphics, addressing the complex challenge of establishing point-to-point correspondences among multiple 3D shapes. By employing a shape graph attention network, DcMatch captures the manifold structure of shape collections, resulting in a more expressive shared latent space. This innovative approach not only enhances the accuracy of shape-to-universe correspondences but also introduces a dual-level consistency mechanism that aligns mappings in both spatial and spectral domains. Extensive experiments validate its effectiveness, demonstrating consistent superiority over previous state-of-the-art methods across diverse scenarios. As the demand for precise 3D shape analysis grows, the implications of this research extend beyond theoretical advancements, potentially influencing applications in areas such as augmented reality, robotics, and digital content creation.