Splat-based Gradient-domain Fusion for Seamless View Transition

Dongyoung Choi, Jaemin Cho, Woohyun Kang, Hyunho Ha, James Tompkin, Min H. Kim

Keywords: novel view synthesis, Gaussian Splatting, Gradient-domain fusion

TL;DR: We introduce Splat-based Gradient-domain Fusion that leverages dense stereo initialization and virtual-view gradient blending to achieve seamless view transitions with smoother colors and more robust synthesis.

Abstract: In sparse novel view synthesis with few input views and wide baselines, existing methods often fail due to weak geometric correspondences and view-dependent color inconsistencies. Splatting-based approaches can produce plausible results near training views, but they frequently overfit and struggle to maintain smooth, realistic appearance transitions in novel viewpoints. We introduce a splat-based gradient-domain fusion method that addresses these limitations. Our approach first establishes reliable dense geometry via two-view stereo for stable initialization. We then generate intermediate virtual views by reprojecting input images, which provide reference gradient fields for gradient-domain fusion. By blending these gradients, our method transfers low-frequency, view-dependent colors to the rendered Gaussians, producing seamless appearance transitions across views. Extensive experiments show that our approach consistently outperforms state-of-the-art sparse Gaussian splatting methods, delivering robust and perceptually plausible view synthesis. A comprehensive user study further confirms that our results are perceptually preferred, with significantly smoother and more realistic color transitions than existing methods.

Submission Number: 277