3D Superquadric Splatting

Gaussian Splatting has proven to be an effective algo- rithm for novel view synthesis and 3D reconstruction from multi-view images. However, the underlying volumetric primitive—the ellipsoidal Gaussian—has limited expressive capabilities, leading to difficulties in 3D modelling (espe- cially geometry such as edges, corners, and high curva- ture). To address this limitation, in this paper, we introduce Superquadric Splats (SQS), an extended class of volumet- ric primitives, as a superset of Gaussian Splats, to model more detailed geometry. We treat superquadrics as vol- umetric distance functions rather than level-set surfaces. A non-trivial differentiable rendering pipeline is developed to support this. Extensive experimental analysis on mul- tiple datasets validates the effectiveness of the proposed SQS approach, showing both enhanced visual and geo- metric performance compared to Gaussian-based splatting (with more than 1dB in PSNR and prominent geometric im- provement). Project page can be found at: https://daniel- macswayne.github.io/3DSQS/