MultiShade
收藏资源简介:
MultiShade是由南洋理工大学等机构构建的合成数据集,包含丰富多样的物体形状、材质和光照条件,用于提升单目法线估计模型在复杂场景下的鲁棒性。该数据集通过模拟不同光照角度下的物体表面反射特性,生成具有几何敏感性的 shading sequence 作为中间表示,其数据规模未明确说明但覆盖了比前人工作更广泛的材质-光照组合。数据创建过程采用计算机图形学技术合成逼真的物体表面反射特性,并通过物理渲染引擎生成多光照条件下的训练样本。该数据集主要应用于计算机视觉领域的3D几何重建任务,旨在解决现有单目法线估计方法存在的3D错位问题,即表面法线视觉合理但与真实几何细节不一致的行业难题。
MultiShade is a synthetic dataset developed by Nanyang Technological University and other institutions. It includes a diverse range of object shapes, materials, and lighting conditions, intended to enhance the robustness of monocular normal estimation models in complex scenarios. This dataset simulates the surface reflection characteristics of objects under varying lighting angles, generating geometry-sensitive shading sequences as intermediate representations. While the exact scale of the dataset has not been explicitly specified, it covers a broader set of material-lighting combinations compared to prior works. The dataset creation process leverages computer graphics technologies to synthesize realistic object surface reflection properties, and generates training samples under multiple lighting conditions using physical rendering engines. Primarily applied to 3D geometric reconstruction tasks in the field of computer vision, this dataset aims to address the 3D misalignment issue prevalent in existing monocular normal estimation methods — a long-standing industry challenge where surface normals appear visually plausible but conflict with real geometric details.

- 1Monocular Normal Estimation via Shading Sequence Estimation南洋理工大学; 字节跳动; 浙江大学; 上海财经大学 · 2026年



