MegaSynth
收藏资源简介:
MegaSynth是由德克萨斯大学奥斯汀分校和Adobe研究院等机构联合创建的一个非语义合成的3D场景数据集,包含700,000个场景,远超现有的真实数据集DL3DV。该数据集通过程序化生成,去除了复杂的语义信息,仅保留基本的场景空间结构和几何原语,从而实现了高效的大规模数据生成。数据集的创建过程包括生成场景平面图、几何和纹理的随机化以及光照的随机化,确保了数据的多样性和复杂性。MegaSynth主要用于训练大型的3D场景重建模型,旨在解决现有数据集规模小、多样性不足的问题,提升模型在多视角图像下的重建能力。
MegaSynth is a non-semantic synthetic 3D scene dataset jointly created by institutions including The University of Texas at Austin and Adobe Research. It contains 700,000 scenes, far exceeding the scale of the existing real-world dataset DL3DV. Generated programmatically, this dataset removes complex semantic information and only retains basic scene spatial structures and geometric primitives, enabling efficient large-scale data generation. The dataset creation process covers generating scene floor plans, randomizing geometry, textures, and lighting, which guarantees the diversity and complexity of the data. MegaSynth is mainly used for training large-scale 3D scene reconstruction models, aiming to address the problems of small scale and insufficient diversity in existing datasets, and improve the model's reconstruction capability under multi-view images.
MegaSynth 数据集概述
数据集名称
MegaSynth
数据集描述
MegaSynth 是一个用于大规模 3D 场景重建的数据集,通过合成数据进行训练。该数据集包含 700K 场景,旨在通过非语义的多视图重建实现可扩展的训练。
关键词
- 3D 场景重建
- 大规模重建模型
- 合成数据
相关链接
作者信息
- Hanwen Jiang<sup>1</sup>
- Zexiang Xu<sup>2</sup>
- Desai Xie<sup>3</sup>
- Ziwen Chen<sup>4</sup>
- Haian Jin<sup>5</sup>
- Fujun Luan<sup>2</sup>
- Zhixin Shu<sup>2</sup>
- Kai Zhang<sup>2</sup>
- Sai Bi<sup>2</sup>
- Xin Sun<sup>2</sup>
- Jiuxiang Gu<sup>2</sup>
- Qixing Huang<sup>1</sup>
- Georgios Pavlakos<sup>1</sup>
- Hao Tan<sup>2</sup>
机构
- <sup>1</sup>UT Austin
- <sup>2</sup>Adobe Research
- <sup>3</sup>Stony Brook University
- <sup>4</sup>Oregon State University
- <sup>5</sup>Cornell University




