Dora-Bench
收藏资源简介:
Dora-Bench是一个用于评估3D形状变分自编码器(VAE)重建质量的基准测试数据集,由香港科技大学、字节跳动Seed、LightIllusions和清华大学联合开发。该数据集包含来自多个公开数据集的3D形状,并根据几何复杂度将其分为四个等级:较少细节、中等细节、丰富细节和非常丰富细节。Dora-Bench通过引入Sharp Normal Error (SNE)度量标准,专注于评估几何细节的重建精度,从而提供了比传统随机采样方法更严格的评估框架。该数据集旨在解决3D形状重建中的几何细节丢失问题,推动3D内容生成领域的研究。
Dora-Bench is a benchmark dataset for evaluating the reconstruction quality of 3D shape variational autoencoders (VAEs), jointly developed by The Hong Kong University of Science and Technology, ByteDance Seed, LightIllusions, and Tsinghua University. This dataset contains 3D shapes sourced from multiple public datasets, and is categorized into four levels based on geometric complexity: low detail, medium detail, high detail, and ultra-high detail. Dora-Bench employs the Sharp Normal Error (SNE) metric to evaluate the reconstruction accuracy of geometric details, thereby providing a more stringent evaluation framework than traditional random sampling-based methods. This dataset aims to address the problem of geometric detail loss in 3D shape reconstruction, and advance research in the field of 3D content generation.
Dora: Sampling and Benchmarking for 3D Shape Variational Auto-Encoders
数据集概述
Dora 是一个用于3D形状变分自编码器(Variational Auto-Encoders, VAEs)的采样和基准测试的数据集。该数据集旨在为3D形状生成和重建任务提供标准化的评估框架。
作者信息
- Rui Chen<sup>1,2</sup>
- Jianfeng Zhang<sup>2*</sup>
- Yixun Liang<sup>1,3</sup>
- Guan Luo<sup>2,4</sup>
- Weiyu Li<sup>1,3</sup>
- Jiarui Liu<sup>1,3</sup>
- Xiu Li<sup>2</sup>
- Xiaoxiao Long<sup>1,3</sup>
- Jiashi Feng<sup>2</sup>
- Ping Tan<sup>1,3*</sup>
通讯作者
- Jianfeng Zhang 和 Ping Tan 为通讯作者。
所属机构
- 香港科技大学(The Hong Kong University of Science and Technology)
- 字节跳动种子(Bytedance Seed)
- LightIllusions
- 清华大学(Tsinghua University)
相关资源




