Rethinking 3D Noise: Learning 3D-Aware Video Priors via Optimization-Free Morphological Perturbations
This work proposes 3D Morphological Perturbations as an optimization-free regularizer that preserves spatial consistency in 3D scene representations and treats each Gaussian as a fundamental building block and applies perturbations across its morphological parameter space via scale, rotation, and pruning.
Onat Şahin, Mohammad Altillawi, George Eskandar et al.
· 0 citations