摘要
神经隐式表面重建技术为三维重建带来了新的契机,然而现有方法存在采样效率低、训练时间长等局限。为此,提出一种基于动态采样范围的优化方案以提升重建的质量与效率。该方案聚焦采样策略,通过占据栅格引导动态生成不断缩小的核心采样区域,结合分层体积采样策略和采样点正则化约束,引导采样点更接近物体表面。实验结果显示,该方案在DTU数据集15个场景下的评价指标均优于基线模型,峰值信噪比为29.79,倒角距离为0.68,与传统方法 NeuS相比分别提升了6.39%和13.24%;训练时间为20 min,相较NeuS方法加速了24倍。基于动态采样范围的神经隐式表面重建优化方案可有效提升重建质量与效率,为三维重建技术的发展提供了新思路。
Neural implicit surface reconstruction technology has brought new opportunities to the field of 3D reconstruction,but existing methods have limitations such as low sampling efficiency and long training time.Therefore,an optimization scheme based on dynamic sampling range is proposed to improve the quality and efficiency of reconstruction.This scheme focuses on the sampling strategy,which dynamically generates a shrinking core sampling area by occupying the grid,and combines the layered volume sampling strategy and sampling point regularization constraints to guide the sampling points closer to the surface of the object.The experimental results show that the evaluation metrics of this scheme are superior to the baseline model in all 15 scenarios of the DTU dataset,with a peak signal-to-noise ratio of 29.79 and a chamfer distance of 0.68.Compared with the traditional method NeuS,it has improved by 6.39%and 13.24%,respectively;The training time is 20 minutes,which is 24 times faster than the NeuS method.The neural implicit surface reconstruction optimization scheme based on dynamic sampling range can effectively improve the reconstruction quality and efficiency,providing new ideas for the development of 3D reconstruction technology.
作者
钟娟
谭诗瀚
梁书凝
陈杨仁
王录涛
ZHONG Juan;TAN Shihan;LIANG Shuning;CHEN Yangren;WANG Lutao(School of Computer Science,Chengdu University of Information Technology,Chengdu 610225,China)
出处
《软件导刊》
2025年第10期88-96,共9页
Software Guide
基金
四川省科技计划项目(2023YFG0304)
四川省重大科技专项(2022ZDZX0008)。
关键词
三维重建
神经隐式表面重建
符号距离函数
占据栅格
分层体积采样
3D reconstruction
neural implicit surface reconstruction
signed distance function
occupancy grid
hierarchical volume sampling
