Deformable Beta Splatting

3D Gaussian Splatting (3DGS) has advanced radiance field reconstruction byenabling real-time rendering. However, its reliance on Gaussian kernels forgeometry and low-order Spherical Harmonics (SH) for color encoding limits itsability to capture complex geometries and diverse colors. We introduceDeformable Beta Splatting (DBS), a deformable and compact approach thatenhances both geometry and color representation. DBS replaces Gaussian kernelswith deformable Beta Kernels, which offer bounded support and adaptivefrequency control to capture fine geometric details with higher fidelity whileachieving better memory efficiency. In addition, we extended the Beta Kernel tocolor encoding, which facilitates improved representation of diffuse andspecular components, yielding superior results compared to SH-based methods.Furthermore, Unlike prior densification techniques that depend on Gaussianproperties, we mathematically prove that adjusting regularized opacity aloneensures distribution-preserved Markov chain Monte Carlo (MCMC), independent ofthe splatting kernel type. Experimental results demonstrate that DBS achievesstate-of-the-art visual quality while utilizing only 45% of the parameters andrendering 1.5x faster than 3DGS-MCMC, highlighting the superior performance ofDBS for real-time radiance field rendering. Interactive demonstrations andsource code are available on our project website:https://rongliu-leo.github.io/beta-splatting/.