Multi-view 3D reconstruction speeds up with adaptive Voronoi cells
CVTSplat: Multi-view Reconstruction Using an Adaptive Centroidal Voronoi Tessellation
Computer Vision and Pattern RecognitionGraphics
Summary
Reconstructing 3D shapes from multiple images usually takes a lot of time because of how space is divided for processing. The authors propose using a special kind of division called adaptive centroidal Voronoi tessellations, which keep the cells well-shaped and adapt their size to focus more on important areas. This approach allows faster computations using graphics hardware and achieves similar accuracy to the best current methods, but trains much faster on a single GPU.
What this means in practice
- •For computer vision developers: Create faster 3D shape reconstruction pipelines from multi-view images for applications like mapping or augmented reality.
- •For graphics software engineers: Develop GPU-accelerated tools for improved real-time 3D surface reconstruction with efficient partitioning and rendering of geometry.
Authors
Briac Toussaint, Diego Thomas, Jean-Sébastien Franco, Edmond Boyer, Christian Theobalt
Abstract
Voronoi tessellations have recently emerged as an adaptive, space-partitioning discretization for shape and appearance reconstruction from multi-view images. However, the irregular cells they produce make the repeated reconstruction of the tessellation during optimization costly. We therefore explore adaptive centroidal Voronoi tessellations (CVTs), which extend standard CVTs with non-uniform cell sizes while preserving the regular, well-shaped cells that make them efficient to compute. In CVTSplat, cell size is modulated by the evolving signed distance field, concentrating sites near the zero level set while keeping cells nearly centroidal. This regularity lets us specialize the whole pipeline, with GPU algorithms for nearest-neighbor queries and approximate Voronoi construction, and a rasterizer tailored to small, nearly isotropic cells. Our Voronoi construction is 5 to 10 times faster than general-purpose alternatives in this regime. On DTU and BlendedMVS, CVTSplat reaches geometric accuracy competitive with the best Gaussian-splatting surface methods while training in about 2.5 minutes on a single GPU.