Learning View-Dependent Splatting Kernels
2026-05-25 • Graphics
GraphicsComputer Vision and Pattern Recognition
AI summaryⓘ
The authors propose a new method that helps computers create better 3D images from different viewpoints. They use a shape called an ellipsoid combined with a small set of learned data to make 2D shapes that change depending on the view. These shapes are created through a neural network system that learns to produce smooth, radially symmetric kernels. Their method performs well on standard tests and can also be used to improve 2D image representation. Overall, the authors show a way to efficiently and accurately generate new views of 3D scenes.
3D view synthesissplattingellipsoid2D kernelMahalanobis distanceprojection networkdecoderneural networksrepresentation efficiencyvolumetric primitive
Authors
Huakeng Ding, Zhanpeng Liu, Fan Pei, Kun Zhou, Hongzhi Wu
Abstract
We present a differentiable framework to automatically learn view-dependent 2D kernels in a splatting-based pipeline to improve reconstruction quality and representation efficiency for novel 3D view synthesis. Our volumetric primitive is defined as a bounding ellipsoid and a 3D-kernel latent vector. We first learn a projection network to output a 2D-kernel latent, taking the attributes of the ellipsoid and the 3D-kernel latent as input. Next, the result is sent to a decoder to produce a radially symmetric 2D kernel in terms of Mahalanobis distance, bounded by the projected ellipsoid. The neural networks along with per-primitive attributes are jointly optimized. The effectiveness of our approach is demonstrated on standard benchmarks, comparing favorably against state-of-the-art techniques on both analytical and learned kernels. Finally, we extend the idea to learn general 2D kernels for 2D splatting as well as image representation.