基于采样优化的边缘辐射算法研究
Research on edge radiation algorithm based on sampling optimization
神经辐射场（NeRF）通过深度学习隐式建模光线辐射场，在低内存下仍可渲染高保真的新颖视图，在三维重建领域得到从而广泛认可。本文则针对NeRF在大规模场景渲染中边缘模糊的问题，提出了一种结合视锥优化、上下界优化及压缩投影的采样方式，有效提升了边缘区域的清晰度和整体渲染的精度。根据实验数据，优化后的NeRF在大规模场景的渲染质量上取得了显著提升，PSNR平均上升了超过3dB， SSIM平均上升了0.193，以及LIPIPS分别下降了0.159，总体实现了更为精确的渲染效果。
Neural radiation Field (NeRF) implicitly models light radiation field through deep learning, rendering high-fidelity novel views in low memory, and has been widely recognized in the field of 3D reconstruction. In order to solve the problem of edge blur in large-scale scene rendering by NeRF, a sampling method combining cone optimization, upper and lower bound optimization and compressed projection is proposed in this paper, which effectively improves the sharpness of edge region and the accuracy of overall rendering. According to the experimental data, the optimized NeRF has significantly improved the rendering quality of large-scale scenes, the average PSNR has increased by more than 3dB, the average SSIM has increased by 0.193, and the LIPIPS has decreased by 0.159 respectively, and the overall rendering effect has been more accurate.
韩利红、李晓乐、陈梓豪
电子技术应用
神经辐射场三维重建辐射适应性采样优化人工神经网络
Nerve radiation fieldThree-dimensional reconstructionRadiation adaptabilitySampling optimizationArtificial neural network
韩利红,李晓乐,陈梓豪.基于采样优化的边缘辐射算法研究[EB/OL].(2024-05-29)[2025-08-11].http://www.paper.edu.cn/releasepaper/content/202405-152.点此复制
评论