基于图神经网络与优化采样策略的隐式表面重建方法

Implicit Surface Reconstruction Method Based on Graph Neural Network and Optimized Sampling Strategy

针对体渲染技术在神经隐式表面重建中常见的细节模糊与局部信息缺失等问题,提出了一种改进的隐式表面重建方法。首先,采用图神经网络(GNN)对不同目标视图进行特征提取,并将这些特征图作为监督信息,引导重建过程。其次,引入一种基于场景几何和光照特征的优化蒙特卡洛路径追踪技术,通过自适应重要性采样策略,优先采样贡献最大的光线路径。最后,利用Omnidata预训练模型提取深度信息和法线信息,对重建过程进行额外的约束。结果表明,与现有技术相比,所提方法在提高表面重建的准确度和视图渲染效果方面表现优异。

Aiming at the common problems of blurred details and missing local information in neural implicit surface reconstruction using volume rendering technology,this paper proposes an improved implicit surface reconstruction method.First,a graph neural network(GNN)is used to extract features from different target views,and these feature maps are used as supervision information to guide the reconstruction process.Secondly,an optimized Monte Carlo path tracing technology based on scene geometry and lighting characteristics is also introduced.Through an adaptive importance sampling strategy,the light path with the greatest contribution is prioritized for sampling.Finally,the Omnidata pre-trained model is used to extract depth information and normal information to impose additional constraints on the reconstruction process.The results show that compared with existing technologies,this method performs well in improving the accuracy of surface reconstruction and the effect of view rendering.