ShapeShifter: 3D Variations Using Multiscale and Sparse Point-Voxel Diffusion

• DOI: 10.48550/arxiv.2502.02187
• Publication Date: 2025-02-04
• Authors: Nissim Maruani; Yifan Wang; Matthew Fisher; Pierre Alliez; Mathieu Desbrun
• Keywords: FOS: Computer and information sciences; Artificial Intelligence (cs.AI); Computer Science - Artificial Intelligence; Computer Vision and Pattern Recognition (cs.CV); Computer Science - Computer Vision and Pattern Recognition

This paper proposes ShapeShifter, a new 3D generative model that learns to synthesize shape variations based on single reference model. While methods for objects have recently attracted much attention, current techniques often lack geometric details and/or require long training times and large resources. Our approach remedies these issues by combining sparse voxel grids point, normal, color sampling within multiscale neural architecture can be trained efficiently in parallel. We show our resulting better capture the fine of their original input handle more general types surfaces than previous SDF-based methods. Moreover, we offer interactive generation variants, allowing human control design loop if needed.