SuperFlex Introduces Bending and Tapering for Improved Superquadric Point Cloud Decomposition (ECCV 2026)

CSProfKGD · x · 2026-08-15

SuperFlex, accepted to ECCV 2026, advances superquadric reconstruction and decomposition by introducing deformation parameters like bending and tapering. This significantly boosts the representation of complex geometries. A novel loss formulation and optimization scheme allow for robust, direct prediction from noisy and incomplete real-world point clouds, overcoming the limitations of previous methods requiring full data. Evaluations on ShapeNet show SuperFlex significantly outperforms existing learning-based and optimization-based methods in metrics like IoU.

Original post →

More from Research

Research channel →