计算机科学
人工智能
正规化(语言学)
曲面重建
算法
迭代重建
计算机视觉
高斯过程
高斯分布
模式识别(心理学)
曲面(拓扑)
信号重构
数学
图像处理
稳健性(进化)
趋同(经济学)
特征提取
自适应滤波器
信号处理
图像分割
迭代法
噪声数据
缩小
作者
Xiao Ren,Yu Liu,Ning An,Jian Cheng,Xin Qiao,Youfu Li,He Kong
标识
DOI:10.1109/tpami.2026.3751295
摘要
Recently, 3D Gaussian Splatting (3DGS) has emerged as an efficient representation for novel view synthesis, offering fast differentiable rendering and high visual fidelity, and has increasingly been extended to surface reconstruction. However, the relationship between depth information and reconstruction error remains insufficiently understood, limiting the principled design of geometric optimization strategies. To address this issue, we propose DepthGS, a depth-guided Gaussian splatting framework. We derive theoretical relations between depth discrepancies and reconstruction error under single-view and multi-view conditions. For single-view optimization, we propose a depth-difference-guided adaptive normal regularization strategy that modulates the normal constraint strength using a geometric reliability weight computed from local depth differences. For multi-view optimization, we introduce a cross-view depth consistency regularization strategy that enforces consistency between neighboring-view depth maps and the corresponding depth predictions transformed from the current view, thereby promoting multi-view geometric consistency. Extensive experiments on DTU, Mip-NeRF360, and Tanks and Temples demonstrate that DepthGS achieves competitive performance in both surface reconstruction and novel view synthesis.
科研通智能强力驱动
Strongly Powered by AbleSci AI