Objaverse
收藏资源简介:
Objaverse数据集由香港科技大学(广州)的研究团队创建,包含200,000个合成3D资产。该数据集通过改变金属性和粗糙度(范围从0到1,步长为0.1)来渲染2D图像和材质映射。数据集的创建过程涉及从Objaverse中采样3D资产,并使用20,000个自然场景的环境映射来提供光照。该数据集主要用于计算机视觉和图形学中的渲染和逆渲染任务,旨在解决从图像中分解出几何、材质和光照信息的问题。
The Objaverse dataset, developed by the research team from The Hong Kong University of Science and Technology (Guangzhou), comprises 200,000 synthetic 3D assets. This dataset generates 2D images and material maps by tuning metallicness and roughness values, which span from 0 to 1 with a step size of 0.1. The dataset creation workflow entails sampling 3D assets from Objaverse and employing 20,000 environment maps from natural scenes to deliver consistent illumination. Primarily applied to rendering and inverse rendering tasks in computer vision and graphics, this dataset targets solving the challenge of decomposing geometric, material, and illumination information from input images.

- 1Uni-Renderer: Unifying Rendering and Inverse Rendering Via Dual Stream Diffusion香港科技大学(广州) · 2024年



