[论文] JanusMesh: Fast and Zero-Shot 3D Visual Illusion Generation via Cross-...

论文概要

研究领域: CV 作者: Siang-Ling Zhang, Huai-Hsun Cheng, Tsung-Ju Yang 发布时间: 2025-06-20 arXiv: 2506.16809

中文摘要

创造三维视觉幻觉——单一三维网格从不同观看角度呈现完全不同语义——是一项迷人却具挑战之任务。现有基于优化之方法缓慢，且易产生过饱和色彩；而朴素拼接方法则难创建几何连贯物体，导致可见不自然接缝与语义泄漏。

本文提出一快速、无需训练之框架，用于文本驱动三维视觉幻觉生成。方法将生成解耦为两阶段：首先，一跨空间双分支降噪过程动态将三维潜在变量解码至体素空间，进行CLIP引导之朝向对齐与SDF融合，确保无缝几何融合；其次，一视点条件纹理合成模块将视点特定二维扩散先验投影并聚合至融合几何上。广泛实验显示，该方法仅需3-5分钟即可生成高度真实之双语义三维幻觉，在几何完整性、语义可识别性与效率上显著优于现有方法。

原文摘要

Creating 3D visual illusions, a single 3D mesh that reveals entirely different semantics from various viewing angles, is a fascinating but tough challenge. Existing optimization-based methods are slow and can produce oversaturated colors. In contrast, naive stitching approaches fail to produce geometrically coherent objects. This results in visible unnatural seams and semantic leaks. In this paper, we present a fast and training-free framework for generating text-driven 3D visual illusions. Our approach decouples the generation into two stages. First, we propose a cross-space dual-branch denoising process. This process dynamically decodes 3D latents into voxel space for CLIP-guided orientation alignment and Signed Distance Field (SDF) blending, which ensures seamless geometric fusion. Seco...

--- *自动采集于 2026-06-20*

#论文 #arXiv #CV #小凯