[论文] UniSHARP: Universal Sharp Monocular View Synthesis

论文概要

研究领域: CV
作者: Meixi Song, Dizhe Zhang, Hao Ren
发布时间: 2025-06-11
arXiv: 2506.08646

中文摘要

本研究致力于将流行的真实感视图合成方法SHARP扩展至通用单目渲染场景，覆盖从传统透视相机到广角、鱼眼及全向全景等多种相机系统。为克服SHARP对针孔模型的固有假设，核心思路是将各类图像统一对齐到全向潜空间中。我们提出UniSHARP，在特征空间与高斯空间中同时执行隐式对齐。具体而言，高斯基元沿光线和径向距离排列于基于光线的通用表征中，同时从UniK3D风格编码器提取的2D语义与3D空间特征被联合解码以生成完整的高斯云。为全面评估方法性能，我们构建了覆盖多种场景与成像系统的基准测试集，并按视场角(FoV)分层以实现细粒度的通用单目渲染评估。大量实验表明UniSHARP显著优于现有方法。

原文摘要

In this work, we focus on extending SHARP, the popular photorealistic view synthesis method, for universal monocular rendering across a continuum of camera systems, from conventional perspective cameras to wide-field-of-view, fisheye and omnidirectional panoramic settings. To overcome the pinhole-specific assumptions of SHARP, our key idea is to align various images in a unified omnidirectional latent space. Thus, we propose UniSHARP, which performs implicit alignment in both feature and Gaussian spaces. Specifically, Gaussian primitives are arranged along rays and radial distances in a ray-based universal representation, while 2D semantic and 3D spatial features extracted from UniK3D-inspired encoders are jointly decoded to generate the complete Gaussian cloud. To comprehensively evaluate...

自动采集于 2026-06-09

#论文 #arXiv #CV #小凯