Three-dimensional(3D)shape registration is a challenging problem,especially for shapes under non-rigid transformations.In this paper,a 3D non-rigid shape registration method is proposed,called balanced functional maps...Three-dimensional(3D)shape registration is a challenging problem,especially for shapes under non-rigid transformations.In this paper,a 3D non-rigid shape registration method is proposed,called balanced functional maps(BFM).The BFM algorithm generalizes the point-based correspondence to functions.By choosing the Laplace-Beltrami eigenfunctions as the function basis,the transformations between shapes can be represented by the functional map(FM)matrix.In addition,many constraints on shape registration,such as the feature descriptor,keypoint,and salient region correspondence,can be formulated linearly using the matrix.By bi-directionally searching for the nearest neighbors of points’indicator functions in the function space,the point-based correspondence can be derived from FMs.We conducted several experiments on the Topology and Orchestration Specification for Cloud Applications(TOSCA)dataset and the Shape Completion and Animation of People(SCAPE)dataset.Experimental results show that the proposed BFM algorithm is effective and has superior performance than the state-of-the-art methods on both datasets.展开更多
This study investigates the impact of the invariance of feature vectors for partial-to-partial point set registration under translation and rotation of input point sets,particularly in the realm of techniques based on...This study investigates the impact of the invariance of feature vectors for partial-to-partial point set registration under translation and rotation of input point sets,particularly in the realm of techniques based on deep learning and Gaussian mixture models(GMMs).We reveal both theoretical and practical problems associated with such deeplearning-based registration methods using GMMs,with a particular focus on the limitations of DeepGMR,a pioneering study in this line,to the partial-topartial point set registration.Our primary goal is to uncover the causes behind such methods and propose a comprehensible solution for that.To address this,we introduce an attention-based reference point shifting(ARPS)layer,which robustly identifies a common reference point of two partial point sets,thereby acquiring transformation-invariant features.The ARPS layer employs a well-studied attention module to find a common reference point rather than the overlap region.Owing to this,it significantly enhances the performance of DeepGMR and its recent variant,UGMMReg.Furthermore,these extension models outperform even prior deep learning methods using attention blocks and Transformer to extract the overlap region or common reference points.We believe these findings provide deeper insights into registration methods using deep learning and GMMs.Our source code and datasets are available at https://github.com/tatsy/DGRM-ARPS.git.展开更多
基金the China Scholarship Council under Grant No.201406070059.
文摘Three-dimensional(3D)shape registration is a challenging problem,especially for shapes under non-rigid transformations.In this paper,a 3D non-rigid shape registration method is proposed,called balanced functional maps(BFM).The BFM algorithm generalizes the point-based correspondence to functions.By choosing the Laplace-Beltrami eigenfunctions as the function basis,the transformations between shapes can be represented by the functional map(FM)matrix.In addition,many constraints on shape registration,such as the feature descriptor,keypoint,and salient region correspondence,can be formulated linearly using the matrix.By bi-directionally searching for the nearest neighbors of points’indicator functions in the function space,the point-based correspondence can be derived from FMs.We conducted several experiments on the Topology and Orchestration Specification for Cloud Applications(TOSCA)dataset and the Shape Completion and Animation of People(SCAPE)dataset.Experimental results show that the proposed BFM algorithm is effective and has superior performance than the state-of-the-art methods on both datasets.
基金supported by a JSPS Grantin-Aid for Early-Career Scientists(JP22K17907).
文摘This study investigates the impact of the invariance of feature vectors for partial-to-partial point set registration under translation and rotation of input point sets,particularly in the realm of techniques based on deep learning and Gaussian mixture models(GMMs).We reveal both theoretical and practical problems associated with such deeplearning-based registration methods using GMMs,with a particular focus on the limitations of DeepGMR,a pioneering study in this line,to the partial-topartial point set registration.Our primary goal is to uncover the causes behind such methods and propose a comprehensible solution for that.To address this,we introduce an attention-based reference point shifting(ARPS)layer,which robustly identifies a common reference point of two partial point sets,thereby acquiring transformation-invariant features.The ARPS layer employs a well-studied attention module to find a common reference point rather than the overlap region.Owing to this,it significantly enhances the performance of DeepGMR and its recent variant,UGMMReg.Furthermore,these extension models outperform even prior deep learning methods using attention blocks and Transformer to extract the overlap region or common reference points.We believe these findings provide deeper insights into registration methods using deep learning and GMMs.Our source code and datasets are available at https://github.com/tatsy/DGRM-ARPS.git.
