摘要
针对实时蒙特卡洛去噪算法效果不佳,且在遮挡情境下会在去噪结果中引入重影、伪影等问题,提出了一种面向遮挡的双边网格实时蒙特卡洛去噪算法。设计添加了可变形卷积的引导图预测网络,以提升去噪效果。引入了双运动向量,以正确指示遮挡情境下的前后帧间像素对应关系,实现遮挡情境下的可靠时域去噪。重新设计了损失函数,旨在增强对历史信息的复用并保障去噪结果的时域稳定性。实验结果表明:该算法在没有增加太多时间开销的情况下,提升了实时去噪效果,实现了遮挡情境下的时域可靠去噪。
To address the issues of suboptimal performance in real-time Monte Carlo denoising algorithms,particularly the introduction of ghosting and artifacts in occlusion scenarios,an occlusion-aware bilateral grid-based real-time Monte Carlo denoising algorithm was proposed.A guided map prediction network incorporating deformable convolution was designed and incorporated to enhance the denoising effect.Dual motion vectors were introduced to accurately indicate the pixel correspondence between preceding and subsequent frames in occlusion scenarios,enabling reliable temporal denoising under such conditions.The loss function was redesigned to strengthen the reuse of historical information and ensure the temporal stability of the denoising results.Experimental results demonstrate that the proposed method improves real-time denoising quality without significant computational overhead,achieving reliable temporal denoising under occlusion scenarios.
作者
李京
胡明月
巩海鑫
熊风光
LI Jing;HU Ming-yue;GONG Hai-xin;XIONG Feng-guang(College of International Education,North University of China,Taiyuan 030051,China;School of Computer Science and Technology,North University of China,Taiyuan 030051,China;Shanxi Key Laboratory of Machine Vision&Virtual Reality,North University of China,Taiyuan 030051,China;Shanxi Province’s Vision Information Processing and Intelligent Robot Engineering Research Center,North University of China,Taiyuan 030051,China)
出处
《计算机工程与设计》
北大核心
2025年第12期3611-3619,共9页
Computer Engineering and Design
基金
国家自然科学基金项目(62272426)
山西省科技重大专项计划“揭榜挂帅”基金项目(202201150401021)
山西省自然科学基金项目(202203021212138、202303021211153、202203021222027)
山西省科技战略研究专项基金项目(202404030401014)。
关键词
实时蒙特卡洛去噪
渲染
深度学习
时域去噪
双运动向量
面向遮挡
可变形卷积
real-time Monte Carlo denoising
rendering
deep learning
temporal denoising
dual motion vectors
occlusionaware
deformable convolution
