Search papers, labs, and topics across Lattice.
This paper introduces TruncGradGS, a novel approach to enhance 3D Gaussian Splatting by addressing the gradient vanishing issue through a piecewise truncated gradient formulation. The method significantly improves optimization stability and robustness, leading to better scene reconstruction from visual input, regardless of initialization method. Experimental results show consistent performance improvements in both static and dynamic Gaussian Splatting scenarios, alongside the introduction of a new dataset for evaluating dynamic scenes.
TruncGradGS tackles the gradient vanishing problem in 3D Gaussian Splatting, leading to superior scene reconstructions across various initialization methods.
3D Gaussian Splatting has become a de facto scene representation for novel view synthesis, yet robustly learning 3D Gaussian primitives from visual input remains challenging. Standard optimization relies on gradient-based updates, but a common issue is the gradient vanishing phenomenon: a pixel far from a Gaussian primitive often has diminishing gradient magnitudes to influence primitive attributes, resulting in suboptimal scene reconstruction. In this paper, we propose a method to address gradient vanishing with a piecewise truncated gradient formulation that improves the optimization stability and robustness to initializations. We show that our method consistently improves 3D Gaussian Splatting with random and COLMAP initializations while being generalizable across static and dynamic Gaussian Splatting. As a by-product, we also examine the limitations of current benchmarks for dynamic scenes, and introduce a novel dataset for benchmarking dynamic Gaussian Splatting using synthetic 3D scenes. We demonstrate the effectiveness of our method in both static and dynamic settings for the public benchmarks and our proposed dataset.
