Surface Correspondence

Surface correspondence focuses on establishing point-wise mappings between the surfaces of 3D shapes, enabling comparisons and transformations across different objects or instances of the same object. Current research emphasizes developing robust and efficient algorithms, often leveraging deep learning architectures like Vision Transformers and neural radiance fields (NeRFs), to predict these correspondences directly from images or point clouds, even without explicit segmentation or manual annotation. This work has significant implications for various fields, including medical image analysis, robotics (e.g., fixture calibration), and computer vision applications like object pose estimation and person re-identification, by facilitating tasks requiring accurate shape understanding and manipulation.