UniG3D
收藏资源简介:
UniG3D是由商汤科技研究院和上海人工智能实验室共同创建的大规模3D对象生成数据集。该数据集通过一个通用数据转换管道,将Objaverse和ShapeNet数据集中的原始3D模型转换为包含文本、图像、点云和网格的全面多模态数据表示。UniG3D旨在解决现有3D对象数据集在文本质量、多模态数据表示完整性和数据集规模方面的不足。数据集创建过程中,利用了渲染引擎和多模态模型来确保文本信息的丰富性和数据表示的全面性。UniG3D的应用领域广泛,包括虚拟现实、自动驾驶、元宇宙、游戏和机器人技术,旨在通过高质量的3D对象生成技术推动这些领域的发展。
UniG3D is a large-scale 3D object generation dataset jointly developed by SenseTime Research and Shanghai AI Laboratory. This dataset employs a universal data conversion pipeline to transform raw 3D models sourced from the Objaverse and ShapeNet datasets into comprehensive multimodal data representations covering text, images, point clouds and meshes. UniG3D aims to address the limitations of existing 3D object datasets in terms of text quality, completeness of multimodal data representations and dataset scale. During the dataset construction process, rendering engines and multimodal models are utilized to ensure the richness of textual information and the comprehensiveness of the data representations. UniG3D has broad application scenarios, including Virtual Reality (VR), autonomous driving, Metaverse, gaming and robotics, and seeks to advance the development of these fields via high-quality 3D object generation technologies.




