Computer Graphics Forum Computer Graphics and Computer-Aided Design

Camera Splatting for Continuous View Optimization

Gahye Lee, Hyomin Kim, Gwangjin Joo, Jooeun Son, Hyejeong Yoon, Seungyong Lee
Wiley · 2026-11-14 · DOI 10.1111/cgf.70625

Abstract

Abstract Given limited camera budgets, identifying informative views is essential for effectively reconstructing a radiance field to achieve high‐quality novel view synthesis. These views should adaptively sample the radiance field based on the geometry and view‐dependent appearance of the scene. Existing methods typically optimize views by selecting from a predefined camera set or adopting a greedy next‐best‐view strategy, limiting the flexibility and accuracy of the optimization. In this paper, we propose a novel continuous view optimization framework, called Camera Splatting , which enables joint optimization of multiple views. Our key idea is to represent each camera as a specialized 3D Gaussian, called camera splat , allowing for gradient‐based optimization within a continuous camera parameter space. To evaluate view optimality, we introduce virtual cameras, called point cameras placed at sampled points in the scene, which render camera splats and provide feedback on their spatial positions and orientations. The resulting optimized views effectively capture the view‐dependent appearance of the scene at the sampled points. Integrated into a Gaussian splatting pipeline, our approach can efficiently and simultaneously optimize a large number of views. Quantitative and qualitative evaluations demonstrate the effectiveness of the views optimized by Camera Splatting for novel view synthesis under a limited camera budget.

Read at publisher →